Expression of MXI1, a Myc Antagonist, Is Regulated by Sp1 and AP2*

MXI1, a member of the MADfamily of Myc antagonists, encodes a transcription factor whose expression must be tightly regulated to maintain normal cell growth and differentiation. To more closely investigate the transcriptional regulation of the human MXI1 gene, we have cloned and characterized the MXI1 promoter. After clarification of the 5′- and 3′-untranslated regions of the cDNA (indicating that the true length of the MXI1 transcript is 2643 base pairs), we identified two transcription initiation sites. We subsequently isolated the MXI1 promoter, which is GC-rich and lacks a TATA box. Although it contains at least six potential initiator sequences, functional studies indicate the proximal two initiator sequences in combination with nearby Sp1 and MED-1 sites together account for virtually all promoter activity. We also demonstrate thatMXI1 promoter activity is repressed by high levels of AP2. These studies provide further insight into the complex regulatory mechanisms governing MXI1 gene expression and its role in cellular differentiation and tumor suppression.

The MYC family of nuclear proto-oncogenes plays a central role in normal cell function, including cell growth, differentiation, and apoptosis (1,2). While the exact role of Myc proteins remains uncertain, expression of the c-MYC gene enhances expression of a number of growth-promoting genes, while inhibiting genes associated with differentiation and growth arrest (3,4). Exogenous expression of c-MYC has been shown in vitro to result in malignant transformation of cells (5,6), and dysregulated c-MYC expression has been identified in a number of human malignancies (7,8).
The c-Myc protein is a basic helix-loop-helix leucine zipper transcription factor that heterodimerizes with a partner protein, Max; this duplex can then interact with target DNA sequences to modulate transcription. Additional basic helixloop-helix leucine zipper proteins able to heterodimerize with Max have been identified, including Mxi1, Mad1, Mad3, Mad4, and Mnt/Rox (9 -13). Unlike Myc, Mxi1 (Max Interactor 1) does not stimulate transcription of target genes but rather interacts with a putative transcriptional repressor, mSin3 (14), which then recruits a histone deacetylase complex mediating transcriptional repression (15,16). A regulatory model can be envisioned in which Myc-Max complexes activate transcription and stimulate cell proliferation, while Mxi1-Max complexes negatively regulate these actions, promoting cellular differentiation. Further studies of MXI1 have supported this model, showing up-regulation of MXI1 mRNA levels in differentiating cells (17)(18)(19). In addition, MXI1 expression vectors are able to suppress MYC-associated neoplastic transformation in rat embryo fibroblast cells (20). We have previously shown that ectopic MXI1 expression results in decreased growth of glioblastoma cells in vitro (21), while DU145 prostate cancer cells infected with an MXI1-expressing adenovirus likewise demonstrate decreased cell proliferation. 1 Finally, homozygous MXI1 knockout mice exhibit an increased susceptibility to tumorigenesis (22). Thus, MXI1 is a putative tumor suppressor gene.
In support of this notion, the MXI1 gene has been localized to chromosome 10q24 -25 (Refs. [23][24][25], a region demonstrating deletions or rearrangements in 60 -97% of human glioblastomas (26,27) and up to 30% of human prostate cancers (28,29). Although loss of heterozygosity for MXI1 is seen in a substantial fraction of glioblastoma tumors (64%), no MXI1 coding sequence mutations have been seen in these tumors (21). Furthermore, a majority of studies have failed to demonstrate MXI1 mutations in prostate tumors (30 -33). Since Mxi1 acts as a Myc antagonist, reduced expression of the MXI1 gene might contribute to development or progression of malignancy. However, none of these studies has analyzed the regulatory regions of the MXI1 gene for mutations or alterations, since little was previously known about the promoter region and transcriptional regulation of the gene. Understanding this regulation may help clarify the role of MXI1 in tumor suppression.
In order to investigate transcriptional regulation of the MXI1 gene, we have cloned and characterized the MXI1 promoter and regulatory regions. After noting potential cloning artifacts in the previously published MXI1 cDNA, we have clarified the true 5Ј-and 3Ј-untranslated regions and identified at least two transcription initiation sites. Subsequent analyses of the GCrich, TATA-less MXI1 promoter delineate a number of critical promoter elements, including two consensus initiator elements (Inrs), 2 a MED-1 site, and Sp1 sites, all of which positively contribute to promoter activity. Finally, we demonstrate that the tissue-regulated and developmentally regulated transcription factor AP2 acts as a negative regulator of the MXI1 promoter.
was carried out in the presence of Superscript reverse transcriptase (20 units/reaction; Life Technologies, Inc.) at 48°C for 4 h. Products were analyzed on a 7 M urea, 8% polyacrylamide gel in parallel with the products of double-stranded sequencing reactions using the same two labeled primers as in the primer extension reactions, according to the manufacturer's instructions for the Taq DNA Sequencing Kit for Standard Sequencing (Roche).
Construction of Chimeric Luciferase Plasmids-1084 base pairs (bp) of genomic sequence preceding the MXI1 ATG was cloned upstream of the promoterless luciferase gene in the XhoI/NcoI sites of the pGL3-Basic vector (Promega) and denoted p3B1084. As a control, a luciferase plasmid was constructed with the 1084-bp insert in the reverse orientation (p3B1084Rev) by digesting the 1084-bp fragment with NcoI, filling in with Klenow fragment of DNA polymerase I, and then digesting with XhoI; this insert was ligated into pGL3-Basic digested with MluI (filled)/XhoI.
A series of 5Ј unidirectional nested deletions of the promoter was generated using the Erase-a-Base system (Promega) with p3B1084 doubly digested with KpnI/NheI. The resultant plasmids were sequenced to determine the extent of the deletion and named according to the length of the promoter fragment (p3B708, p3B533, p3B421, p3B374, and p3B297). The p3B308 construct was created by PCR, using p3B1084 as the template, the sense primer F-308 (5Ј-CCGCTCGAGCG-GCGGGACTACATTTCCCAGGG) incorporating an XhoI site (underlined), and the antisense GL2 primer (Promega), complementary to luciferase gene sequences downstream of the MXI1 promoter insert. The PCR product was digested with XhoI/NcoI and ligated into pGL3-Basic/XhoI/NcoI. Plasmid preparations of all constructs were purified with a plasmid preparation kit (QIAGEN).
Site-specific Mutagenesis-Full-length promoter constructs containing point mutations of potential Inrs or the MED-1 site were generated using p3B1084 as template, mutated oligonucleotide primer pairs (Table I), and the RVp3 and GLp2 primers (Promega). Briefly, a 5Ј-mutated promoter fragment was amplified using RVp3 and the desired mutated antisense primer, while an overlapping 3Ј-mutated promoter fragment was amplified using GLp2 and the corresponding mutated sense primer. The 5Ј and 3Ј PCR products were heated to 94°C and cooled slowly to allow annealing of the overlapping fragments. RVp3 and GLp2 were then used to amplify the full-length promoter containing the appropriate mutation. All PCRs were performed with GC-Melt TM and Advantage®-GC polymerase mix (CLONTECH). The full-length PCR fragments were digested with XhoI/NcoI and ligated into pGL3-Basic/ XhoI/NcoI. Although two different mutations of Inr 2 were constructed, they gave similar results on all assays, so their data were combined.
The p3BDel construct containing a deletion of the promoter sequence from Ϫ535 to Ϫ498 was generated by digesting p3B1084 with either AccI (at Ϫ535) or DraIII (at Ϫ498), treating with Klenow, and then digesting with XhoI or NcoI, respectively. These two fragments were ligated into pGL3-Basic/XhoI/NcoI. Combination mutants were constructed either by similar PCR strategies as above, using singly mutated p3B1084 DNA as template, or by utilizing restriction enzyme sites between mutations to isolate and then recombine mutant promoter fragments in desired combinations. Relevant regions of all wild-type and mutant constructs were sequenced to confirm the presence of the mutation(s).
Transient Transfections and Luciferase Assays-K562 cells in log phase were washed with ice-cold Opti-MEM (Life Technologies, Inc.) and resuspended in Opti-MEM at a concentration of 5 ϫ 10 7 /ml. 1 ϫ 10 7 cells were preincubated with 5 g of the indicated plasmid and 50 ng of pRL-TK Renilla vector at room temperature for 10 min. Cells were transfected by electroporation with a BTX Electro Cell Manipulator 600

5Ј-GTGCGGGACTgggTTTCCCAGGGGGCAG
3Ј-CCGGGGGCCGCGAgtAtacCGGGGAGGGGGCACG at 350 V, 650 microfarads, and 13 ohms. Cells were then incubated at room temperature for 15 min and transferred to 3 ml of warmed RPMI 1640 medium with additives. The transfected cells were maintained in 5% CO 2 at 37°C for 17 h, at which time they were harvested, and both luciferase and Renilla activity were measured using the Dual-Luciferase TM assay system (Promega). Luciferase activity values were normalized to the Renilla activities for each sample (L/R). All reactions were performed in triplicate and repeated at least three times; results are expressed as the mean Ϯ S.E. For AP2 assays in HepG2 cells, HepG2 cells in log phase were trypsinized, washed with ice-cold Opti-MEM, and then resuspended at a concentration of 2.5 ϫ 10 7 /ml. 0.5 ϫ 10 7 cells were preincubated with 5 g of the promoter construct p3B1084, 1 g of pRL-TK Renilla vector, and the indicated amounts of a pCMX-PL1/AP2␣ expression vector (gift of R. Buettner) at room temperature for 10 min. Cells were transfected by electroporation at 150 V, 1700 microfarads, and 186 ohms. Cells were then incubated at room temperature for 30 min and transferred to 3 ml of warmed modified Eagle's medium with additives. The transfected cells were maintained in 5% CO 2 at 37°C for 48 h, at which time they were harvested (with scraping) and assayed as above.
Gel Shifts-Double-stranded oligonucleotide probes (I-IV, as illustrated in Fig. 6) were synthesized by heating primer pairs in the presence of GC-Melt TM to 95°C for 10 min and then cooling slowly to room temperature. Probe I/MMED was designed with a 6-bp mutation of the MED-1 site from GCTCCC to CATATG but was otherwise identical to Probe I. DNA-protein reactions were performed in gel mobility shift buffer (10 mM Tris-HCl (pH 7.4), 80 mM NaCl, 1 mM dithiothreitol, and 5% glycerol), with 5 g of K562 nuclear extract (Santa Cruz Biotechnology, Inc., Santa Cruz, CA), 0.25 g of herring sperm DNA, and 0.5 g of poly(dI-dC). Reactions with purified protein contained 1 footprint unit recombinant Sp1 or AP2 (Promega) with 1 g of herring sperm DNA and 0.5 g of poly(dI-dC). Where indicated, unlabeled competitor Sp1 or AP2 oligonucleotide (Promega) was included. Reactions were incubated at room temperature for 15 min prior to and again after adding 33 P-labeled probe. For certain reactions, Sp1 or AP2 antibody (Santa Cruz Biotechnology) was then added, and incubation continued at room temperature for 15 min. The DNA-protein complexes were analyzed on a 5% polyacrylamide gel run at 16°C in 0.25ϫ TBE buffer at 200 V for approximately 2 h.

RESULTS
Clarification of the MXI1 3Ј-Untranslated Region (UTR)-We previously obtained and partially sequenced a genomic clone encompassing the MXI1 gene (34). Comparison of the genomic sequence with the published MXI1 cDNA sequence (35,36) revealed several discrepancies. First, the MXI1 genomic sequence contains a 16-bp poly(A) tract corresponding to the site of cDNA termination, which may have resulted in the construction of an artificially truncated cDNA. There is, however, a consensus polyadenylation sequence (AATAAA) 102-bp downstream of the terminal A in the genomic poly(A) tract (Fig. 1A). To confirm the extent of the MXI1 3Ј-UTR, we performed 3Ј-RACE with a poly(dT) synthesis primer. PCR amplification of this cDNA using an MXI1-specific oligonucleotide primer (Revrep 146) detected two bands, a lower molecular weight one corresponding to the genomic poly(A) tract, and a second more prominent band of ϳ275 bp (Fig. 1B). DNA sequencing of this second PCR product confirmed that the 3Ј-UTR extends an additional 126 bp downstream of the genomic poly(A) tract.
Clarification of the 5Ј-UTR-Comparison of the MXI1 5Ј-UTR with the corresponding genomic sequence revealed an abrupt loss of homology of the genomic sequence with the cDNA just upstream of the ATG translation start site. Analysis of this region revealed that nucleotides 68 -174 in the published 5Ј-UTR are an exact inverted repeat of the final 107 nucleotides of the cDNA (Fig. 1A), suggesting that the 5Ј inverted repeat sequence may have resulted from a cloning artifact. This was particularly likely, since the genomic sequence upstream of the MXI1 translation start site is extremely GCrich. Furthermore, previous attempts to detect the presence of this inverted sequence by Southern blotting consistently demonstrated only a single band corresponding to the 3Ј repeat (34). To clarify the MXI1 5Ј-UTR, we employed the 5Ј-RACE technique with K562 and HL60 RNA and several nested antisense oligonucleotides corresponding to recognized MXI1 exon 2 and exon 1 cDNA sequences. The resulting 5Ј-RACE products confirmed that the true MXI1 cDNA sequence extends at least 254 bp upstream of the ATG; although this sequence loses homology with the published cDNA sequence at the region of the inverted repeat, it corresponds exactly to the MXI1 genomic sequence (data not shown).
Identification of the MXI1 Transcription Initiation Site(s)-Initial attempts to identify the transcription initiation site(s) of the MXI1 gene using 5Ј-RACE proved difficult due to the extremely GC-rich nature of this region. However, primer extension analysis using two sequence-specific reverse primers revealed at least two potential start sites, as well as a number of faint bands over a several hundred-bp range (Fig. 2). Simultaneous sequencing reactions determined that the two strongest primer extension bands correspond to a G at Ϫ307 (major site), and a C at Ϫ274 (minor site).
Characterization of the 5Ј Regulatory Region-In order to evaluate MXI1 promoter and regulatory regions, we isolated and sequenced a 1084-bp fragment of genomic DNA immediately upstream of the MXI1 translation start site (GenBank TM accession no. AF148881). Sequence analysis revealed neither TATA nor CAAT boxes. However, the region immediately proximal to the ATG is extremely GC-rich, with an 80% GC content between Ϫ1 and Ϫ300. Just upstream of this GC-rich region are six consensus sequences for Inr sites (YYAN(A/T)YY) over a 230-bp range (Fig. 3). A search for potential transcription factor binding sites revealed seven Sp1 (GGGCGG) (37) and numerous AP2 consensus sequences (CCC(A/C)N(G/C)(G/ C)(G/C) or GCCNNNGGC) (38, 39) clustered in the region between Ϫ1 and Ϫ410, an area also containing the first two potential Inrs. Of note, one of the transcription initiation sites identified by primer extension (at Ϫ307) is located 4 bases upstream of the first consensus Inr, while the second start site (at Ϫ274) lies within a nearby Sp1 binding sequence. Just downstream of the transcription initiation sites is a consensus MED-1 site (multiple start site element downstream, GCTCC(C/G)) (40) at Ϫ250 (see below). Further upstream is a polypurine tract at Ϫ491, as well as several E boxes and a potential p53 binding site (not shown).
Functional Analysis of the MXI1 Promoter-Potential promoter activity of the MXI1 5Ј-flanking region was investigated using a luciferase reporter vector with 1084 bp of MXI1 genomic sequence (p3B1084). Transient transfection of this vector into K562 cells, which constitutively express MXI1, demonstrated an 11-fold increase in promoter activity relative to the pGL3-Basic empty vector (Fig. 4). This promoter activity was largely abolished using a construct containing the 1084-bp MXI1 fragment cloned into pGL3-Basic in reverse orientation (p3B1084Rev). Several additional cell lines, including U87MG and U373MG glioblastoma, HL60 myelomonocytic, and U937 promonocytic cell lines, also demonstrated increased promoter activity when transfected with the p3B1084 promoter construct (data not shown). Although the degree of promoter activity varied by cell line, in each case it was essentially not seen with the p3B1084Rev construct.
To delineate the minimal promoter region as well as critical regulatory regions, we constructed a series of nested 5Ј deletions and compared their promoter activities in K562 cells with that of p3B1084 (100% activity). The results of this analysis (Fig. 4) revealed a 47% increase in promoter activity with p3B708, suggesting the loss of a negative regulatory element within that 376-bp deleted region. Further 5Ј deletion from Ϫ708 to Ϫ421, with sequential loss of potential Inrs 3-6, did not significantly alter promoter activity, with p3B421 still having 130% activity. However, deletion of an additional 47 bp of 5Ј sequence to Ϫ374 resulted in a decrease in promoter activity to 47%, a nearly two-thirds reduction relative to the p3B421 construct. Notably, this 47-bp promoter region contains the Inr 2 consensus sequence at Ϫ400. Progressive deletion of promoter sequence between Inr 2 and Inr 1 (p3B308) resulted in loss of promoter activity to only 20% of that of the full-length construct. Finally, the p3B297 construct, in which Inr 1 is also deleted, showed little luciferase activity above base line. These results suggest that the first 308 bp upstream of the MXI1 translation start site contain the minimal sequences required for promoter activity.
The Effect of Initiator Element Mutations on Promoter Activity-To further evaluate the effect of each putative Inr on overall promoter activity, we mutated or deleted potential Inr sequences and assayed their promoter activities in K562 cells relative to wild-type p3B1084 (Fig. 5). The construct p3BDel (deletion of Ϫ535 to Ϫ498, including Inrs 4 -6) demonstrated a minimal decrease in MXI1 promoter activity to 82%. Point mutation of Inr 3 or Inr 2 likewise resulted in minimal decreases in promoter activity (to 80 or 73% activity, respectively). The construct in which the Inr 1 consensus site was mutated demonstrated the largest single reduction in promoter activity, having only 46% activity relative to wild-type p3B1084.
Since none of these individual Inr mutations completely abolished MXI1 promoter activity, we next constructed combination mutants. Although deletion of Inrs 4 -6 or mutation of Inr 3 caused slight decreases in promoter activity, a promoter construct containing both the deletion of Inrs 4 -6 and mutation of Inr 3 (p3BM3/Del) showed a 64% increase in promoter activity over the wild-type p3B1084 construct (Fig. 5). On the other hand, combining mutations of Inrs 1 and 2 (p3BM1/2) resulted in a dramatic reduction in promoter activity to 22%, suggesting that these two Inr sequences are critical MXI1 promoter elements. Finally, a promoter construct in which Inrs 4 -6 were deleted and Inrs 1-3 were mutated (p3BM1/2/3/Del) still maintained 34% of promoter activity, a slight increase over that of p3BM1/2 but consistent with the fact that p3BM3/Del alone resulted in an increase in promoter activity.
The Role of MED-1 and Sp1 Sites in the MXI1 Promoter-The fact that no combination of Inr mutations completely eliminated promoter activity suggested the presence of additional promoter elements capable of initiating transcription, albeit at a low level. Sp1 is a ubiquitous transcriptional activator that may initiate transcription, either independently or in concert with Inrs in other TATA-less promoters (41,42). Evaluation of the MXI1 promoter revealed 7 consensus Sp1 binding sites in the 325-bp, GC-rich region just upstream of the start codon. Four of these Sp1 sites (Sp1-D, -E, -F, and -G) are clustered in an 80-bp region surrounding Inr 1; one of these four (Sp1-F) contains the minor transcription initiation site identified by primer extension (Fig. 6). In addition to the potential Sp1 sites, there is also a consensus MED-1 site at Ϫ250; this sequence was first described in a class of TATA-less promoters with multiple transcription initiation sites (40). Located just down- stream of the start site window, it contributes to utilization of multiple start sites. We speculated that the residual promoter activity seen despite mutation of Inrs 1 and 2 might be a result of Sp1-mediated transcription initiation at minor start sites, regulated in part through the MED-1 site.
To investigate whether Sp1 could bind to any of the consensus Sp1 sequences in the MXI1 promoter, we performed gel shift assays. In Fig. 7A, a double-stranded DNA probe including MXI1 sequence from Ϫ227 to Ϫ262 (probe I; see Fig. 6), containing the Sp1-D site and the partially overlapping MED-1

FIG. 4. Measurement of MXI1 promoter activity by transient transfection in K562 cells.
Luciferase activity of p3B1084 is shown on the right relative to both the promoterless luciferase reporter (p3B) and a construct with the 1084-bp promoter fragment in reverse orientation (p3B1084Rev). Additional promoter constructs are illustrated on the left, with their luciferase activity relative to p3B1084 (100%) shown on the right. Construct names represent the length of the promoter fragment in bp. The Inrs are indicated by the black boxes (Inrs 4 and 5 are shown as one box, since their sequences overlap), and the major transcription start site at Ϫ307 is represented by the arrow. Luciferase activities are normalized to cotransfected Renilla activities, with each experiment being performed in triplicate a minimum of three times.

FIG. 5. Effect of Inr mutations on
MXI1 promoter activity. Promoter constructs containing mutations or deletions of potential Inrs, either singly or in combination, are illustrated on the left, with black boxes representing intact Inr sequences and hatched boxes indicating the mutated/deleted Inrs. The names of each construct indicate the number of the Inr mutated, with the exception of p3BDel, in which Inrs 4 -6 were deleted as a group. These constructs were transiently transfected into K562 cells, and their resultant corrected luciferase activities, relative to p3B1084 (100%), are shown on the right. All experiments were performed in triplicate and repeated a minimum of three times. site, was used as a labeled probe. Both pure Sp1 protein (lane 2) and K562 nuclear extract (lane 3) bound to the probe; these similarly shifted complexes were specifically competed away with unlabeled Sp1 oligonucleotide (lane 4) and were supershifted with anti-Sp1 antibody (lane 5). Additionally, the lanes with K562 nuclear extract (lanes 3-5) demonstrated a faster migrating complex (arrow 1) that may be due to the binding of unknown proteins to the MED-1 site. Mutation of the MED-1 site, which also partially mutated the Sp1 site from GGGCGG to GGGCCA, did not affect Sp1 binding (lanes 6 -10). However, the faster migrating complex was dramatically diminished, while a slightly higher band was amplified (arrow 2), consistent with altered binding to the mutated MED-1 site. Similar gel shift experiments with probes containing MXI1 sequence from Ϫ258 to Ϫ287 (probe II) and Ϫ285 to Ϫ325 (probe III) showed that Sp1 was capable of binding specifically to each of these probes, both as a pure protein and in K562 nuclear extracts (data not shown). These results support the hypothesis that Sp1 binding to sites in the MXI1 promoter may play an important role in its regulation. In addition, the MED-1 sequence at Ϫ250 appears to support protein-DNA interactions independent of Sp1 binding nearby.
We next began to assess the functional contribution of the MED-1 and Sp1 sites to MXI1 promoter activity. Despite mul-tiple attempts to prepare a reporter construct with a MED-1/ Sp1-D mutation by PCR, every effort resulted in concomitant mutations in one or more nearby Sp1 sites. One of these constructs, p3BMMED/Sp1, contained a point mutation in the Sp1-F site that encompassed one of the transcription initiation sites, in addition to the original MED-1 mutation. When transfected into K562 cells, this construct showed no significant change in promoter activity compared with p3B1084 (Fig. 7B). We then combined these mutations with mutations of both Inrs 1 and 2. While p3BM1/2 still retained 22% of the full promoter activity seen with p3B1084, the addition of the MED-1/Sp1 mutations (p3BM1/2/MED/Sp1) effectively abolished all promoter activity. Thus, it appears that the MED-1 site and at least one Sp1 site, in conjunction with the proximal two Inrs, are functionally significant MXI1 promoter elements, accounting for nearly all of the promoter activity in K562 cells.
To determine the effect of overexpression of Sp1 on the MXI1 promoter, we cotransfected an Sp1 expression vector in addition to p3B1084 into K562 cells. These experiments showed no consistent change in promoter activity with the overexpression of Sp1 (data not shown). Since Sp1 is ubiquitously expressed in mammalian cells, we postulated that the promoter construct might have been saturated with endogenous Sp1. This cotransfection analysis was then performed in Drosophila SL2 cells, which lack Sp1. Transfection of the p3B1084 MXI1 promoter construct alone demonstrated no significant promoter activity; cotransfection with the Sp1 expression vector did not alter this lack of activity (data not shown). It would appear, then, that Sp1 alone cannot drive expression of MXI1; additional transcription factors not present in Drosophila cells must be required for MXI1 transcriptional activity.
AP2 Repression of the MXI1 Promoter-The MXI1 promoter includes multiple consensus AP2 transcription factor binding sites, many of which overlap Sp1 sites (Fig. 6). Since AP2 has been implicated in both positive and negative transcriptional regulation of other genes with GC-rich promoters (43,44), we investigated the role of AP2 in the regulation of MXI1 promoter activity. We first performed gel shift experiments with the same labeled MXI1 probes described previously. Probes I, II, and III each contain at least one consensus AP2 binding sequence; Probe IV (Ϫ390 to Ϫ424) also includes an AP2 binding site but no consensus Sp1 sequence. A representative gel shift using probe III, containing overlapping AP2/Sp1 sites, is shown in Fig. 8A. The labeled probe bound purified AP2 protein (lane 4), which was competed away by unlabeled AP2 oligonucleotide (lane 5); it also bound pure Sp1 (lanes 2 and 3). However, when K562 nuclear extracts were used, the probe did not appear to bind AP2. Instead, there was a slower migrating doublet, consistent with Sp1 binding, and a faster migrating triplet of bands. Since this probe contains Inr 1 and one of the transcription start sites, these lower bands may represent transcription initiation-related DNA-protein complexes. Similar gel shift experiments were performed with probes I, II, and IV, all of which bound purified AP2 protein, but did not demonstrate shifted bands consistent with AP2 binding with K562 nuclear extracts. Thus, although purified AP2 can bind to several AP2 consensus sequences in the MXI1 promoter, in the context of K562 nuclear extracts, Sp1 rather than AP2 bound preferentially.
To assess the functional effect of AP2 expression on the MXI1 promoter, we utilized HepG2 human hepatocarcinoma cells, which do not express AP2. The p3B1084 MXI1 promoter construct was cotransfected with varying amounts of an AP2␣ expression vector into HepG2 cells (Fig. 8B). Although there was a slight increase in promoter activity with low amounts of AP2, there was a significant dose-dependent repression of the MXI1 promoter with increasing AP2, suggesting that AP2 behaves as a repressor of MXI1 transcription at high concentrations.

DISCUSSION
MXI1, a candidate tumor suppressor gene that has been implicated in the induction and maintenance of the differentiated state, appears to function in a tightly regulated network with the MYC oncogene in order to maintain normal cell growth and differentiation. Loss of MXI1, as seen in MXI1deficient mice (22), results in increased susceptibility to tumorigenesis, while overexpression of MXI1 in both glioblastoma (21) and prostate tumor cells 1 leads to decreased tumor growth. Understanding the regulation of MXI1 expression should contribute to our insight into the regulation of normal cell growth and differentiation. In this study, therefore, we set out to identify and characterize the regulatory regions of the MXI1 gene.
Published studies had delineated a 2417-bp MXI1 cDNA (9, 36) but an approximately 2800 -3200-bp product on Northern analysis (18). In comparing the cDNA sequence with the corresponding sequence of the genomic clone, we noted several potential reasons for this discrepancy, including an inverted repeat sequence in the 5Ј-UTR. Further investigation of both the 3Ј-and 5Ј-untranslated regions of the cDNA using RACE techniques determined that the full MXI1 cDNA contains an additional 126 bp in the 3Ј-UTR. More significantly, we were able to clarify the 5Ј-UTR, which varies from the previously published cDNA but corresponds exactly to the genomic sequence. These additional 3Ј and 5Ј sequences result in an MXI1 cDNA of 2590 bp. Using primer extension assays, we identified two distinct transcription initiation sites at Ϫ307 and Ϫ274. When the major start site at Ϫ307 is taken into account, the calculated length of the MXI1 cDNA is 2643 bp. Although some of the primer extension assays did reveal additional weaker bands throughout the GC-rich region, it is unclear whether these represent true transcription initiation sites or merely premature pausing of the reverse transcriptase. Since many genes with GC-rich TATA-less promoters have multiple transcription initiation sites (45)(46)(47), it is certainly possible that there are additional MXI1 start sites, creating some variability in the cDNA size.
We next isolated and sequenced a 1084-bp region of genomic sequence immediately upstream of the MXI1 translation start site. Evaluation of this sequence revealed a number of elements consistent with an active promoter. Although there are no  5 and 8). The position of the Sp1-DNA complexes and AP2-DNA complexes are indicated by arrows. B, the effect of AP2 on MXI1 promoter activity was determined by transiently transfecting an AP2-null cell line (HepG2) with 5 g of the chimeric p3B1084 MXI1 promoter construct and increasing amounts of an AP2 expression vector (pCMX-PL1/AP2␣). The resulting corrected luciferase activities are expressed as a percentage of the activity of the p3B1084 construct when transfected with the empty pCMX-PL1 vector (the combined amount of pCMX-PL1/AP2 ϩ empty pCMX-PL1 vector was identical in all samples). These assays were performed in triplicate and repeated a minimum of three times. canonical TATA or CAAT boxes, there are six potential Inrs between Ϫ530 and Ϫ300. TFIID is thought to interact with Inr sequences, recruiting and stabilizing components of the transcription initiation machinery (reviewed in Refs. 48 and 49). In addition to the potential Inrs, the MXI1 promoter region contains multiple Sp1 and AP2 consensus binding sites, a MED-1 site, and a polypurine tract just upstream of Inr 3. Polypurine tracts, which are essential elements in some promoters, may function by altering the structure or stability of the nucleosome, increasing the accessibility of downstream sequences to gene-activating proteins (50). The presence of all these elements clustered over several hundred base pairs upstream of the MXI1 ATG strongly supports the likelihood that this region is a functional promoter.
Subsequent promoter analyses confirmed that the 1084 bp of sequence upstream of the MXI1 ATG has significant promoter activity. This activity was abrogated when the 1084-bp sequence was in the reverse orientation, consistent with the orientation-specific requirement of an active promoter. A series of promoter deletions and Inr mutations demonstrated that significant MXI1 promoter activity requires at least 421 bp of sequence upstream of the ATG. This region includes Inrs 1 and 2, which both contribute to transcriptional activation. Interestingly, the sequences adjacent to and including these two Inrs demonstrate a remarkable 14/17 bp homology with each other. However, mutation of Inr 1 resulted in the greatest single reduction of promoter activity. Further reduction in promoter activity was seen with a construct containing a double mutation of Inrs 1 and 2. Mutation of Inrs 3-6 did not cause significant loss of promoter activity, either individually or in addition to mutations of Inrs 1 and 2, confirming that these four potential Inrs do not contribute significantly to overall promoter activity. Therefore, the residual promoter activity seen despite mutation of Inrs 1 and 2 suggests the ability of additional elements to contribute to transcriptional activation.
One of the sequences potentially contributing to transcriptional activation is the consensus MED-1 site at Ϫ250. This sequence (GCTCC(C/G)) has been described in a series of TATA-less promoters with multiple transcription initiation sites; it is found just downstream of the start site window and contributes to the utilization of multiple start sites (40). When this element was mutated in the P-glycoprotein promoter, transcription was reduced due to decreasing utilization of the downstream start sites. In the MXI1 promoter, gel shift experiments using K562 nuclear extracts demonstrated that a probe with the mutated MED-1 sequence had altered protein-DNA interactions when compared with the nonmutated probe. Although this MED-1 mutation also partially mutated the overlapping Sp1-D site, Sp1 binding to the probe was not affected. Functionally, mutation of the MED-1 site in combination with any of the Sp1 sites did not alter overall promoter activity when Inrs 1 and 2 were intact. However, the residual promoter activity seen despite mutation of Inrs 1 and 2 was abrogated by the addition of the MED-1/Sp1 mutations.
It has been noted that Inr-containing promoters are frequently responsive to specific transcriptional activators such as Sp1, which contributes to basal promoter activity by stabilizing the transcriptional machinery (41,48). Although Sp1 is ubiquitously expressed, it is involved in the transcriptional regulation of a number of tissue-specific and differentiation-dependent genes, including genes regulating neural, epithelial, and hematopoietic cell differentiation (44,51,52). Our promoter analyses support a role for Sp1 in the transcriptional activation of MXI1 for several reasons. First, the MXI1 sequence upstream of the ATG has 7 consensus Sp1 sites between Ϫ105 and Ϫ320, four of which are clustered around Inr 1 and the MED-1 site. Second, one of the transcription initiation sites detected by primer extension localized to the Sp1-F site, just downstream of Inr 1. Third, gel shift experiments demonstrated that Sp1 was capable of binding to each of the four clustered Sp1 sites, even in the presence of a partial mutation of the Sp1-D binding site. Fourth, functional promoter studies demonstrated a loss of promoter activity when sequence containing the Sp1-G site just upstream of Inr 1 was deleted (p3B374 to p3B308). Finally, the MXI1 promoter still retained more than 20% of its original activity despite mutation or deletion of all putative Inrs, suggesting the involvement of additional transcriptional activators such as Sp1. This residual promoter activity was abolished completely by adding mutations of the Sp1-F and MED-1 sites to the Inr 1/2 mutations, confirming a role for these sites in contributing to MXI1 promoter activity in K562 cells.
We noted with special interest the presence of multiple consensus AP2 sequences in the MXI1 5Ј-UTR and promoter region. The product of one of three related genes (AP2␣, AP2␤, or AP2␥), AP2 is a developmentally regulated and tissue-specific transcription factor expressed primarily in neural crest cell and epidermal cell lineages (43,53,54). Although AP2 was initially described as a transcriptional activator, several recent studies have demonstrated its ability to behave as a repressor. In both the human acetylcholinesterase gene and the K3 keratin gene, there are overlapping Sp1 and AP2 sites in the promoter, with Sp1 activating and AP2 repressing gene transcription (44,51). In rabbit corneal epithelial cells stimulated to differentiate, AP2 is dramatically down-regulated, thus altering the Sp1:AP2 ratio and favoring the activation of the K3 keratin gene by Sp1 in differentiating cells (51). Since MXI1 is highly expressed in brain and its promoter also has multiple overlapping Sp1 and AP2 sites, we postulated a role for AP2 in MXI1 transcriptional regulation. Cotransfection of AP2 with the MXI1 promoter construct p3B1084 in HepG2 cells (which lack endogenous AP2) clearly demonstrated a significant repression of promoter activity with increasing levels of AP2. In gel shift assays, both purified AP2 protein and Sp1 protein were able to bind to several MXI1 probes containing overlapping AP2 and Sp1 sites; however, only Sp1 bound in the context of K562 nuclear extracts. It may be that the concentration of AP2 and the relative ratio of AP2 to Sp1 in a particular cell environment determine the efficiency of binding of AP2 to its binding sites in the MXI1 promoter, thus partially controlling the balance between gene activation and repression. In support of this hypothesis, MXI1 promoter activity (with the p3B1084 construct) was lower in all glioblastoma cell lines compared with the K562 cell line, perhaps due to repression by endogenous AP2 in the glioblastoma cells (data not shown). Additional studies of the role of AP2 in MXI1 regulation in neural cell lines should further clarify this issue.
Multiple elements contribute to the basal transcriptional activation of the MXI1 gene, including Inrs 1 and 2, the MED-1 site, and several Sp1 sites. This lack of a single well defined mechanism for transcriptional activation may confer significant advantages and allow for more precise regulation. Although some of the first GC-rich, TATA-less promoters studied were associated with "housekeeping genes," it is now clear that many Inr-based genes are highly tissue-specific or developmentally regulated (47,55). The use of Inrs rather than TATA boxes may allow differential regulation by activators and repressors depending upon a particular cell milieu. For example, certain proteins such as p53 repress transcription from some TATA-containing promoters but not from Inr-based promoters (56). Other transcription factors such as Myc have been shown to activate transcription through an E-box but repress tran-scription through an Inr (57)(58)(59). This type of differential regulation might allow a gradation in gene expression during development or a mechanism by which cells can initiate the cascade of events leading to differentiation.
In addition to the regulatory elements discussed above, the transcriptional regulation of MXI1 expression may also rely on certain structural elements. This is particularly likely with respect to the GC-rich 5Ј-UTR, especially in the region between Ϫ230 and Ϫ300. When analyzing this region, we noted that experimental modifications designed to limit formation of secondary structures were critical, suggesting that this region may lend itself to structural constraints that are difficult to disrupt. In support of this, computer-based RNA structural analyses of these sequences indicate the possible formation of a number of stem-loop structures within this region. Interestingly, all of our MED-1/Sp1 mutant constructs developed concomitant mutations in one or more nearby Sp1 sites, consistent with the formation of favorable stem structures involving Sp1 sites. It may be that this region serves as a negative regulatory element, limiting access to Sp1 and Inr 1 sites. If so, the binding of a protein to the MED-1 site may alter these structural constraints, increasing accessibility of this region to the transcription initiation machinery. On the other hand, the absence of factors able to bind the MED-1 sequence may result in limiting access to the Inr 1 region, favoring transcription initiation from Inr 2.
In summary, we have demonstrated that regulation of the MXI1 promoter is complex and dependent on a number of factors. Multiple Inrs and transcription initiation sites may allow for more precise transcriptional regulation. Involvement of Sp1 and AP2, both of which play a role in neural development, supports a role for MXI1 in the developing nervous system. Finally, structural constraints may impose yet another level of regulation. Ongoing studies should more precisely elucidate the transcriptional regulation of the MXI1 gene and contribute to our understanding of mechanisms of gene regulation.