In Vivo Structure of Two Divergent Promoters at the Human PCNA Locus

Proliferating cell nuclear antigen (PCNA) synthesis is strictly regulated during the cell cycle. To investigate PCNA transcriptional regulation, we have analyzed protein-DNA interactions at the promoter region and in the first intron in quiescent fibroblasts and following serum stimulation. Twenty putative protein-binding sites, distributed in two divergent promoters at the PCNA locus, were identifiedin vivo by genomic footprinting. These elements bind transcription factors continuously throughout the cell cycle with the exception of one E2F consensus site, located in the first intron at position +583. This E2F site becomes strongly occupied 18 h after serum stimulation, implying that an E2F activator complex plays a role in activation of the PCNA gene at the onset of S phase. We detected a 500–600-base pair-long antisense transcript by Northern blot analysis. This RNA has no apparent coding capacity and is constitutively transcribed from a promoter located within the first intron. We suggest that silencing of the PCNA gene is accomplished through base pairing between sense pre-mRNA and antisense RNA. The binding of S phase-specific E2F complexes at the +583 element may help to overcome the negative effect of the antisense transcript, which results in up-regulation of PCNAexpression in proliferating cells.

The proliferating cell nuclear antigen (PCNA) 1 gene product is a nuclear protein, whose synthesis correlates with the proliferative state of the cell. As a cofactor of DNA polymerase ␦ (1, 2), PCNA is involved in DNA replication and repair, as well as cell cycle progression and cell differentiation, and it represents a key gene necessary for the transition of cells from quiescence to S phase. PCNA synthesis is induced in quiescent cells by stimulation with serum, growth factors, or adenovirus infection and fluctuates during the cell cycle (3)(4)(5)(6)(7)(8)(9)(10). Expression of this gene is regulated at both transcriptional and post-transcriptional levels (for review see Refs. 3 and 11).
The levels of PCNA mRNA are controlled in part by elements located in the 5Ј region of the gene (12)(13)(14). The 5Ј-flanking region of the human PCNA gene was first characterized by Travali et al. (15). They showed that a promoter including the sequences up to a PvuII restriction site (located at nt Ϫ395 upstream of the cap site) were sufficient to drive the transcription of a linked reporter gene, regardless of the orientation of the promoter (15,16). Pietrzkowski et al. (14) reported that sequences between Ϫ73 and Ϫ45 of the human PCNA promoter contain a sequence that shows all the characteristics of an enhancer. This enhancer-like structure is able to actively increase the levels of PCNA mRNA, but it does not play a role in the serum regulation of mRNA levels (14). The human PCNA promoter is fully active in serum-deprived murine 3T3 cells, suggesting that either PCNA expression is not regulated at the level of transcription initiation or that sequences other than the 5Ј promoter may be responsible for serum-dependent growth regulation of PCNA (17). It has been suggested that elements in intron 4 (18,19) and in intron 1 (20) also participate in transcriptional regulation of the gene.
In this study, we have analyzed both the 5Ј-flanking region and the first intron of the human PCNA gene to characterize regulatory elements involved in its expression at the onset of S phase. In vivo genomic footprinting has indicated that a key element for cell cycle regulation of the PCNA gene is located in the first intron. We further characterized this regulatory element and suggest a dominant role of an E2F complex as an S phase-specific positive regulator of PCNA expression. The potential role of a newly discovered 500 -600-bp-long antisense PCNA transcript originating from the first intron is also discussed.

MATERIALS AND METHODS
Cell Culture and Cell Synchronization-Normal human foreskin fibroblasts (HF39) were grown in 4% CO 2 at 37°C in maintenance medium consisting of Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. Cells were made quiescent by serum deprivation for either 2 or 14 days and then re-stimulated to proliferate by the addition of fresh Dulbecco's modified Eagle's medium supplemented with 15% fetal calf serum (21). Cell cycle synchronization and DNA synthesis were monitored by flow cytometric analysis using a Becton-Dickinson fluorescence-activated cell sorter scan, as described previously (22).
RNA Analysis-Subconfluent HF39 human fibroblasts were rendered quiescent by incubation in serum-free Dulbecco's modified Eagle's medium for either 2 or 14 days. After addition of 15% fetal calf serum, cells were harvested at the times indicated, and total RNA was isolated by standard procedures (RNAgents, Promega). A 10-or 20-g aliquot of each RNA sample was separated on formaldehyde-agarose gels and transferred to Genescreen membranes (NEN Life Science Products). Northern blots containing approximately 2 g of poly(A) ϩ RNA from human cancer cell lines and human multiple tissues were purchased from CLONTECH and used according to the manufacturer's instructions. Hybridization was performed at 60 -62°C in 0.25 M sodium phosphate, pH 7.2, 7% SDS, 1 mM EDTA, 1% bovine serum albumin, as described (21). Single-stranded probes, specific for the first exon of the PCNA gene, were generated by linear PCR amplification on either strand using appropriate synthetic oligonucleotides. Probes specific for either the human GAPDH or the human ␤-actin gene were included as controls.
The levels of the antisense transcript in nuclear and cytoplasmatic fractions were analyzed by a semiquantitative RT-PCR assay. Human foreskin fibroblasts were lysed on ice as described previously (21), and cell lysates were centrifuged for 10 min at 4°C. The pelleted nuclei and the supernatant were recovered separately and treated with RNAzol B reagent for RNA isolation (Tel-Test, Inc.). First strand cDNAs were synthesized from DNase I-treated RNAs (150 and 300 ng) by use of Moloney-murine leukemia virus reverse transcriptase (Life Technologies, Inc.) and a PCNA sense primer (oligo Ϫ11: GACGCGGCGGCAT-TAAAC, Tm ϭ 56°C). Diluted cDNAs were then subjected to PCR amplification with the sense primer and an antisense primer (oligo ϩ372: GTTCACGCCCATGGCCAG, Tm ϭ 56°C) to generate a 383-bplong fragment. PCR-amplified products were separated on a 1.8% agarose gel and analyzed by Southern blotting with a single-stranded PCR probe, specific for exon 1.
Immunoblot Analysis-Nuclear extracts were prepared from fibroblasts that were serum-starved for 14 days (0 h) and at 12, 18, 24, or 30 h after serum stimulation, as described previously (21,29). Approximately 8 g of protein were resolved on a 12% SDS-polyacrylamide gel and transferred onto Hybond TM enhanced chemiluminescence membrane (Amersham Pharmacia Biotech). The membrane was then incubated with a mouse monoclonal antibody to PCNA (sc-56, Santa Cruz Biotechnology, Santa Cruz, CA), diluted 1:1,200 in phosphate-buffered saline-0.1% Tween 20 (PBS-T). After incubation with horseradish peroxidase-conjugated anti-mouse IgG antibody (1:1,000 dilution) and repeated washes in PBS-T, the signal was detected using an ECL method, as suggested by the manufacturer's instructions (Amersham Pharmacia Biotech).
Genomic Footprinting with Dimethyl Sulfate (DMS)-Immediately after the serum starvation period and at different time points after serum stimulation, subconfluent human fibroblasts (HF39) were treated with 0.2% DMS (Aldrich), and nuclei were isolated as described (21). DNA was isolated from nuclei and cleaved at methylated bases with hot piperidine (23). The chemically cleaved DNA was then amplified by ligation-mediated PCR (LM-PCR), and the sequence ladders were separated on 8% acrylamide, 7 M urea gels in 0.1 M Tris borate EDTA (TBE) and transferred to nylon membranes (24,25). 32 P-labeled single-stranded hybridization probes were synthesized by a PCR-based technique using a strand-specific primer (see below for details) from the appropriate human PCNA primer set (25,26). Naked DNA controls (C, CϩT, GϩA, G reactions) were obtained by in vitro treatment of fibroblast DNA or HeLa DNA with DMS, as described previously (27), and were included on the sequencing gels to provide position markers. Four gene-specific LM-PCR primer sets, selected with the aid of a computer program (28), were used to analyze the promoter region of the human PCNA gene and are listed below. For lower strand analysis, the following primers were used: PCNA-A primer set: A1, TTCACCTCTTTATC-TTGTGACACC, Tm ϭ 52°C; A2, TGACACCTACGAGCGCATCAATT-CTGTAAT, Tm ϭ 64°C; A3, TTGAAAAATAAAGTGCATATTTGCAG-CAGC, Tm ϭ 62°C; PCNA-C primer set: C1, GGACCGGGACCCGAT, Tm ϭ 50°C; C2, GGACCGGGACCCGATCTCCACA, Tm ϭ 66°C; C3, CGATCTCCACATATGCCCGGACT, Tm ϭ 61°C.
Nuclear Extracts and Mobility Shift Assay-Nuclear extracts for DNA binding reactions were prepared from human diploid fibroblasts as described previously (21,29). To inhibit protease activities, phenyl-methylsulfonyl fluoride (final concentration, 0.5 mM) was included in the extraction buffers. DNA binding assays were carried out as described in Ausubel et al. (30) with some modification. A typical binding mixture contained 13 mM Hepes, pH 7.9, 13% glycerol, 64 mM KCl, 0.13 mM EDTA, 0.5 mM MgCl 2 , 0.3 mM phenylmethylsulfonyl fluoride, 0.5 g of salmon sperm DNA, and 6 -8 g of nuclear extracts. DNA⅐protein complexes were resolved by electrophoresis through low ionic strength 6% polyacrylamide gels at 4°C in TAE or TBE buffer (21). A synthetic oligonucleotide spanning the sequence ϩ583 (5Ј-AACCTGCTTTTTCG-CGCCAAAGTCACAAAG) was gel-purified, end labeled with [␥-32 P] ATP and T4 polynucleotide kinase (31), annealed with a 2-fold excess of the cold complementary strand and used as a probe in the binding reactions. The same gel-purified unlabeled oligonucleotide was annealed in molar equivalent quantities with the lower strand and used as a cold competitor.

Cell Cycle Dependence of PCNA Gene Expression in Human
Fibroblasts-Normal human foreskin fibroblasts (strain HF39) were used for serum starvation and cell cycle synchronization. Periods of starvation longer than 48 h (up to 14 days) have previously been shown to drastically improve the degree of cell synchrony after serum stimulation and were preferentially used in our experiments. It has been determined that the longer starvation time does not alter normal cell cycle progression (21). The cell cycle profile was the same as described previously (22).
The induction of the human PCNA gene after serum stimu-

FIG. 1. Expression analysis of human PCNA.
A, normal human foreskin fibroblasts were serum-starved for either 2 or 14 days and then stimulated to re-enter the cell cycle by addition of 15% fetal calf serum. Total RNA was isolated from quiescent cells at various time points after serum stimulation, and a 10-g aliquot was analyzed by Northern blotting. A single-stranded probe specific for the first 100 bp of transcribed sequences of human PCNA mRNA was used. In lanes labeled Fib. and HeLa, total RNA was isolated from unsynchronized fibroblast and HeLa cell cultures, respectively. A probe specific for the GAPDH gene was included as a control. B, immunoblot analysis of PCNA protein in cells that were serum-starved for 14 days and then re-stimulated with serum. Fib., nuclear extracts from unsynchronized fibroblasts. lation was analyzed in synchronized fibroblasts by Northern blot analysis. Low amounts of PCNA mRNA were detected in serum-starved fibroblasts after a starvation time of either 2 or 14 days (Fig. 1A). The levels of PCNA mRNA increased in serum-stimulated cells, reaching a peak of expression at 18 -20 h after re-addition of serum when the majority of the cells have entered S phase (22). Much higher levels of PCNA mRNA were present in unsynchronized cycling HeLa cells than in unsynchronized fibroblasts (Fig. 1A). Western blot analysis showed that the PCNA protein, unlike its RNA, is undetectable in serum-starved cells and in G 1 cells. The protein appears between 18 and 24 h following serum stimulation (Fig. 1B).
In Vivo Footprinting Analysis of the Human PCNA Promoter-To detect potential upstream cell cycle regulatory elements, we investigated protein-DNA interactions in vivo at the promoter of the human PCNA gene in serum-starved fibroblasts and at various time points following serum stimulation. A sensitive genomic footprinting technique was used to identify changes at the single nucleotide level in the binding pattern of transcription factors as the cells progress through the cell cycle. Briefly, human fibroblasts were treated with DMS, a methylating agent that reacts predominantly with guanines at the N-7 position, enabling later cleavage of the modified bases by piperidine. Sequence ladders were then amplified by LM-PCR and analyzed on sequencing gels as described (21). In a situation where transcription factors are bound to the DNA, they will either increase or decrease accessibility of DMS to specific guanines, resulting in a signal of hyperreactivity or protection, respectively. When DMS patterns from quiescent fibroblasts and cells at various time points after serum stimulation are compared with DMS patterns from naked DNA controls, it is possible to detect occupied sites and to identify those elements at which changes in protein-DNA interactions may occur during cell cycle progression. We designed four different LM-PCR primer sets to cover approximately 500-bp of the PCNA promoter (15,32). This sequence included 400 bp of the region upstream of the transcription initiation site (ϩ1) and the first 100-bp of transcribed sequences. First, we analyzed upperstrand sequences surrounding the transcription start site, between nucleotides Ϫ126 and ϩ18 ( Fig. 2A). This region contains clear footprints at a potential Sp1 consensus binding site (5Ј-GGGGGCGGG) located between nucleotides Ϫ130 to Ϫ122 and a footprint at nucleotides Ϫ70 to Ϫ66 (5Ј-GTGGG). These sites are occupied in vivo before serum addition and during all subsequent cell cycle phases, indicating that binding is constitutive. The binding pattern is retained after 14 days of serum starvation, indicating that the longer starvation time does not interfere with normal cell cycle regulation. This region contains also an inverted CCAAT box element at position Ϫ95. The guanines within the CCAAT box are protected from DMS mod- Lanes G, fibroblast DNA treated with DMS in vitro. Lanes exp.1, DNA from fibroblasts that were serum-starved for 2 days and treated with DMS at 0, 5, 10, 15, 20, or 25 h following serum addition. Lanes exp.2, DNA from fibroblasts that were serum-starved for 14 days and then treated with DMS at 0, 6, 12, 18, 24, or 30 h after serum stimulation, Lanes Div. Fib. and Div. HeLa, DNA from DMS-treated unsynchronized fibroblast and HeLa cell cultures, respectively. Open and closed circles at the right side of the gel indicate guanine residues that are hyporeactive and hyperreactive, respectively, to in vivo DMS treatment. Shaded boxes indicate the footprints. A, the sequences cover the upper strand of the PCNA promoter from nt ϩ18 to nt Ϫ126. An arrow indicates the transcription initiation site. Primers PCNA-D1, -D2, and -D3 were used for LM-PCR analysis. B, the sequences cover the lower strand from nt Ϫ34 to nt Ϫ172 and were analyzed with primers PCNA-C1, -C2, and -C3. ification throughout cell cycle progression, indicating that this site is occupied in vivo in a constitutive manner. An additional footprinted area is also visible in the middle of the gel at nucleotide Ϫ53 (5Ј-GTGACGTCG). This element represents a perfect consensus site for the transcription factor ATF (5Ј -TGACGC/TC/AG/A). The adenovirus E1A protein has been shown to activate the PCNA promoter through this proximal ATF element (13,33,34). We also analyzed the opposite strand of the sequences illustrated in Fig. 2A. Fig. 2B shows a genomic footprinting analysis of the lower strand spanning nucleotides Ϫ34 to Ϫ172 relative to the transcription start site. As observed on the opposite strand, there are several Sp1 binding sites located within this region, a CCAAT box element and an ATF binding site. No change in protein-DNA interactions was detected during cell cycle progression along the entire sequence analyzed in Fig. 2, A and B.
We then investigated the binding pattern of sequences further upstream in the 5Ј end of the gene, approximately up to nucleotide Ϫ400. The results are presented in Fig. 3, which shows the same region on the upper (Fig. 3A) and lower ( Fig.  3B) strand, respectively. Two Sp1-like binding elements are visible in position Ϫ185 to Ϫ191 (upper strand) and in position Ϫ167 to Ϫ160 (lower strand), and both are protected from DMS modification in a constitutive manner. A weak footprint can also be detected on both strands, between positions Ϫ240 and Ϫ220. This sequence deviates only two nucleotides from a consensus site for p53 binding (consisting of two copies of the repeat RRRC(A/T)(A/T)GYYY). Previous data demonstrated that p53 can modulate activation of the PCNA gene through this site in response to DNA damage, although such induction is cell specific and strictly dependent on the level of p53 expression (8). This may explain the weak interactions seen at the p53 binding site, together with the fact that DMS has been shown to be a poor footprinting reagent with respect to p53 binding (35).
Genomic Footprinting Analysis of the First Intron of the PCNA Gene-The role of introns in the growth regulation of mRNA levels of the human PCNA gene has been investigated. Alder et al. (20) suggested that intron 1 down-regulates the levels of PCNA mRNA in quiescent cells due to interaction with transcription factors and that its inhibitory effect is alleviated by binding of additional proteins in growing cells. To further elucidate the intron-dependent cell-cycle regulation of PCNA, we analyzed protein-DNA interactions in a 300-bp-long region within the first intron of the gene. According to the footprinting data, at least three potential Sp1 binding sites are located within this region, at positions ϩ471, ϩ513, and ϩ601 (Fig.  4A). These Sp1 consensus sites are all occupied in vivo before serum addition and at all subsequent phases of the cell cycle, suggesting constitutive binding of regulatory factors. Clear footprints were also observed at two inverted CCAAT box elements spanning position ϩ468 to ϩ464 and from position ϩ499 to ϩ495, respectively (visible at the top of the gel in Fig. 4A). It has been suggested previously that intron 1 may exert its negative regulatory effect on the expression of PCNA through the CCAAT box element located at position ϩ468 to ϩ464 (20). We noticed that the guanines within the core motif of the CCAAT elements (both proximal and distal) were protected from DMS modification at all time points, indicating that the sites were occupied in vivo continuously during cell cycle progression. Considering the in vivo footprinting pattern, one may question the involvement of the CCAAT box as a major candidate in PCNA cell cycle regulation, although we cannot rule out the possibility that post-translational modifications of these factors and/or interaction with cofactors may regulate transcription even if binding itself is constitutive.
We only found one position, near nt ϩ583 in the first intron, where the in vivo footprint pattern undergoes a change as a function of the cell cycle. At this sequence, two guanines are weakly protected from DMS modification in quiescent cells and at the beginning of G 1 phase. This protection becomes much stronger around 18 h after serum stimulation, when PCNA mRNA reaches its peak of expression (Fig. 1A) and is retained until the end of G 2 phase (30 h). An identical pattern of footprint was also visible on the opposite strand and was confirmed by PhosphorImager analysis (Fig. 4, C and D). These data indicate that a factor may be bound weakly at this position before serum addition, but upon expression of the gene, a major switch in binding occurs. A late G 1 -S associated factor may bind either to the DNA or to preexisting bound proteins to create a higher binding affinity at this stage, pointing toward the existence of an inducer.
The data from all in vivo footprinting experiments are summarized in Fig. 5. We analyzed up to 400 bp upstream of the transcription start site at the promoter region, but in vivo footprints were found only between Ϫ200 and ϩ1. This is in agreement with results showing that the HpaII promoter, i.e. the promoter including the restriction site HpaII located 210 bp upstream from the cap site, is sufficient for growth regulation of PCNA mRNA levels (11). We detected at least nine, perhaps ten, different binding sites in this region. These include a potential binding site for p53, several Sp1 binding sites, two CCAAT boxes, one perfect consensus element for the ATF factor, and three sites occupied by factors of unknown identity. None of these binding sites showed any change in the footprint-ing pattern during cell cycle progression, suggesting that other sequence elements may be more important for cell cycle regulation of this gene.
An almost identical set of occupied factor binding sites was FIG. 4. Genomic footprinting of the PCNA first intron in human fibroblasts. Lanes marked C, CϩT, GϩA, and G are Maxam-Gilbert control sequences. In vitro DMS-treated "naked" DNA (G lanes) is compared with fibroblasts that were serum-starved for 14 days and treated in vivo with DMS at 0, 6, 12, 18, 24, and 30 h following serum stimulation. Guanine residues reacting differentially after DMS treatment are indicated by circles (E, hyporeactive; q, hyperreactive). Footprints are marked by boxes. The arrows indicate the E2F site that shows a change in the binding pattern. A, sequences from the upper strand spanning from nt ϩ657 to nt ϩ458 were analyzed with primers PCNAin1-4, -5, and -6. B, sequences from the opposite strand, from nt ϩ488 to ϩ639 were analyzed with primers PCNAin1-1, -2, and -3. C and D, PhosphorImager data for the E2F footprint, which changes as a function of the cell cycle. The band intensities of protected guanines were normalized relative to several adjacent G positions that did not change during the cell cycle. The data are for the upper (C) and lower (D) strands. found in the first intron of the gene, between nt ϩ450 and nt ϩ660. Nine, probably ten, different transcription factor binding elements can be recognized, including two inverted CCAAT boxes, several Sp1 elements, a putative p53 binding site, and several possible binding sequences for the transcription factor E2F (Fig. 5). Of the latter, the one in position ϩ583, is particularly interesting, being the only site in the entire sequence analyzed that shows a cell cycle dependence in respect to DMS footprints. We have therefore focused most of our attention on characterization of the ϩ583 element.
Characterization of a Putative E2F Site in the First Intron of the Human PCNA Gene by Gel Mobility Shift Assays-The sequence from nt ϩ579 to nt ϩ590 (5Ј-TTTCGCGCCAAA), which shows footprint changes as a function of the cell cycle, is a perfect consensus motif for the transcription factor E2F. Using gel shift assays, we determined protein binding to this site during cell cycle progression. We noticed that the pattern of protein⅐DNA complexes is very similar to the one obtained using an E2F oligonucleotide from the adenovirus E2 promoter as a probe in electrophoretic mobility shift assays (21). A faster migrating complex was seen in G 0 cells but disappeared 12 h after serum addition. One or two slower migrating complexes also disappeared when the cells entered S phase (18 h) and were replaced by a new complex of different mobility (Fig. 6). All these complexes are highly specific because they are competed by an excess of cold ϩ583 oligonucleotide, but they are only partially competed by an excess of cold Ϫ20 oligonucleotide from the human CDC2 promoter (Fig. 6). This Ϫ20 element, which deviates by at least 1 base pair from known E2F consensus sites, was shown to negatively regulate CDC2 in quiescent cells (21). To gain information about the identity of these complexes, specific antibodies were included in the gel shift reactions. Several antibodies directed against proteins commonly found in E2F complexes were used. These included antibodies against E2F-1, E2F-2, E2F-3, E2F-4, E2F-5, and antibodies against the pocket proteins, retinoblastoma gene product (Rb), p107 and p130 (Figs. 6 -8). Antibodies directed against the E2F-4 protein (sc-512x and sc-1082x, in Figs. 6 and 7B, respectively), and in a separate reaction, antibodies directed against the pocket protein p130 (Fig. 8) supershifted FIG. 5. Summary of the genomic footprinting data. E and q represent guanines that are hypo-and hyper-reactive to DMS in vivo, respectively. Unmarked nucleotides showed no reactivity difference. Putative transcription factor binding sites are indicated by boxes. The asterisk indicates the E2F element that shows a cell cycle-dependent footprint. The horizontal arrow marks the transcription initiation site (ϩ1) and the broken vertical arrow indicates the beginning of intron 1. Sequence changes in HeLa DNA are also shown above the fibroblast DNA sequence, deletions are indicated by a dash, and base changes are indicated by capital letters.
most of the complexes. The predominant form of E2F in G 0 cells contains the E2F-4 protein in association with p130, but a minor supershift was observed also with an anti-E2F-5 antibody (Fig. 7B). Minor supershifts were also observed with an antibody against E2F-3. p130 is still the major pocket protein found in late S and G 2 phases after serum stimulation although p107 complexes are also detectable (Fig. 8). Experiments with deoxycholate, which disrupts protein-protein complexes, showed that the majority of E2F at all cell cycle phases is in the form of higher order complexes (Fig. 7A). Cyclin A was not detected in these complexes (data not shown).
Antisense Transcription from PCNA Intron 1-The E2F site, a potential candidate for the cell cycle regulation of the PCNA gene, is located in an approximately 200-bp-long sequence within intron 1, which has all the characteristics of a functional promoter. To investigate possible mechanisms involved in introndependent regulation of PCNA transcript levels, we performed Northern blot analysis. Hybridization with a radiolabeled sense probe specific for the PCNA first exon revealed the presence of a short divergent (antisense) transcript (500 -600-bp long). This transcript is constitutively expressed in quiescent cells and at different phases of the cell cycle (Fig. 9A). The expression of this antisense transcript is relatively low, considering that a larger amount of total RNA (20 g) is necessary to obtain a signal after an overnight exposure. The antisense transcript is predominantly nuclear, although it is also present in cytoplasmatic fractions, as determined by a semiquantitative RT-PCR assay (Fig. 9B). This short transcript is expressed ubiquitously in a panel of normal human tissues (higher in pancreas with an additional longer transcript) and in cancer cell lines with the exception of promyelocytic leukemia HL-60 (Fig. 10B). Probes specific for the lower strand of the 5Ј end of the gene (between nt Ϫ1178 and nt Ϫ200) failed to hybridize on the Northern blots, suggesting that both the sense and the short antisense transcripts contain the same PCNA exon 1 and that no other gene is transcribed in the opposite orientation upstream of the PCNA promoter (data not shown). With the FIG. 6. Electrophoretic mobility shift assay of the putative E2F site at position ؉583. Nuclear extracts were prepared from serumstarved cells (0 h) and at 12, 18, 24, and 30 h following serum stimulation. A 32 P-labeled double-stranded oligonucleotide spanning the ϩ583 sequence in the PCNA first intron was used as the probe. A 250-fold excess of unlabeled competitor oligonucleotides was used to assess to specificity of the complexes. 2 g of anti-E2F-4 antibody was used to supershift the complexes. FIG. 7. Composition of the complexes binding to the ؉583 sequence located in intron 1. Nuclear extracts were prepared from cells that were serum-starved for 14 days (0 h) and at 12, 18, or 24 h after serum stimulation. A 32 P-labeled, double-stranded oligonucleotide spanning the ϩ583 site was used as the probe. A and B, nuclear extracts were incubated with the respective antibodies against E2F family members. Lane DOC, complexes were treated with 0.6% sodium deoxycholate. exception of thymus and testes, PCNA mRNA is expressed at low levels in most of the normal human tissues analyzed (Fig.  10A). This may be explained by the fact that all these tissues contain asynchronous populations of cells, and therefore the expression pattern resembles the one of dividing fibroblasts, where the majority of the cells are in G 1 phase and PCNA expression is low (Fig. 1A, lane 13). On the other hand, in almost all cancer cell lines analyzed, we detected very high levels of PCNA mRNA, suggesting a high proportion of cells in S phase or a potential lack of transcriptional regulation in malignant cells (Fig. 10A). The level of the antisense transcript is generally higher in normal tissues compared with malignant cells, and the opposite is true for the sense PCNA transcript (Fig. 10). However, there is not always a direct inverse correlation of transcript levels for each individual tissue or cell line.

DISCUSSION
Transcriptional Regulation of the Human PCNA Gene Involves Two Promoters-PCNA is regulated by a combination of mechanisms that act at both a transcriptional and post-transcriptional level (11). It has been reported that introns, in addition to the 5Ј-flanking sequences, play an important role in cell cycle regulation at a transcriptional level. In particular, intron 1 and intron 4 have been proposed to be responsible for serum responsiveness and growth regulation (17,20). We noticed that the sequences spanning intron 4 are very rich in AT and therefore poorly fit as a candidate for a characteristic cell cycle-regulated promoter. In this study, we examined cell cyclespecific binding of transcription factors, concentrating our attention on the analysis of intron 1 together with the upstream promoter region of the human PCNA gene. Potential transcription factor binding sites, identified by DMS footprinting in vivo, are summarized in Fig. 5. At least 20 regulatory elements are equally and almost symmetrically distributed in two face-toface divergent promoters, part of the same PCNA locus and separated by only 400 bp of exon 1. Both these promoters present the typical features of many cell cycle-regulated genes (21,22,36). Besides being TATA-less, they include two CCAAT boxes, E2F-like elements, and a series of GC elements, probably binding sites for Sp1. They also contain two potential consensus binding sites for p53, which may point to a role of PCNA in DNA repair (8,37).
Role of E2F in Cell Cycle Regulation of the Human PCNA Gene-An increasing number of reports document the pivotal role of the transcription factor E2F in coordinating transcription during the mammalian cell cycle, particularly in the induction of specific genes at the G 1 /S transition (38 -41). There are at least 5 putative elements in the entire PCNA sequence analyzed that resemble a binding site for E2F (5Ј-TTTCGCGC).  9. Expression of the antisense transcript. A, Northern blot analysis of the human PCNA gene using a sense probe. The Northern blot contained total RNA (20 g) from quiescent (14-day starvation) and synchronized fibroblasts at the time points indicated. In lane Fib., total RNA was isolated from an unsynchronized fibroblast cell culture. A 23-mer oligonucleotide, spanning from nt ϩ177 to nt ϩ199 (5Ј-GGTC-CAGGGCTCCATCCTCAAGA), was designed to generate a singlestranded PCR probe specific for the antisense transcript. A probe specific for the GAPDH gene was included as a control. B, RT-PCR data for the antisense transcript in nuclear (lanes 1 and 2) and cytoplasmatic (lanes 3 and 4) fractions of fibroblasts. 150 and 300 ng of DNase I-treated RNA were used for RT-PCR amplification (lanes 1 and 3 and  lanes 2 and 4), as described in the text. The first one is located just downstream of the ATF-1 binding site in the promoter region of the gene, at position Ϫ43 to Ϫ34 (5Ј AACGCGGCGC; the base pairs differentially reactive to DMS are underlined). However, the guanines in this sequence are hyperreactive to DMS, which is not a typical feature of in vivo E2F footprints (21,22,42). The other E2F-like sites are located, in the first intron, from nt ϩ482 to ϩ490 (5Ј-TCCGC-GCCTT), from nt ϩ546 to ϩ556 (5Ј-AAAGCCCGCGC), from nt ϩ579 to nt ϩ590 (5Ј-TTTCGCGCCAAA), and from nt ϩ608 to ϩ 616 (5Ј-GGCGGGAAA), respectively. Among all these putative sites, only the element located at positions ϩ579 to ϩ590, which is also a perfect consensus sequence for E2F, shows cell cycle-dependent binding of transcription factors in vivo, suggesting a role of this E2F site in controlling expression of the gene. The involvement of E2F in PCNA regulation, through an intron-associated binding site has been proposed recently. Lee et al. (5) demonstrated that 2.8 kilobases of the PCNA 5Јflanking sequences led to constitutive transcription, but the inclusion of a segment from the first intron of the gene was able to enhance transcription in G 1 /S-enriched nuclei. Furthermore, co-transfection of E2F-1 and DP-1 expression plasmids into human Saos-2 cells can activate transcription of PCNA-CAT constructs that contain an intact intron 1 but not a PCNA intronless reporter gene. Normal CAT activity was re-established by insertion of the ϩ583 E2F sequence from the first intron (5). These results are in agreement with our in vivo data, suggesting a role of E2F as a transactivator of PCNA gene expression through its induced binding in intron 1. Electrophoretic mobility shift assay analysis with antibodies suggests that the protein complex interacting with the ϩ583 element contains mainly E2F-4, but probably not E2F-1. In addition, it contains the pocket protein p130 but not the Rb protein. E2F-1, which is thought to be the major activator of the E2F family in late G 1 /S phase, was not detected by mobility shift assays with two different antibodies used. However, this does not exclude the possibility that E2F-1 may function in vivo to activate the PCNA gene via intron 1 in other cell types; its abundance in human fibroblasts may simply be too low relative to E2F-4 to be detectable. The subcellular localization of E2F-4 is apparently mostly cytoplasmic after S phase entry (43,44), but nevertheless we detected this protein as a major E2F species in nuclear extracts from S and G 2 phase cells as well (Figs. 6 -8). E2F-4 can function as a transcriptional activator when overexpressed (45)(46)(47)(48)(49). Because the mobility shift assays failed to detect any free E2F species in S phase, the precise nature and function of the presumed activating E2F complex remains unknown. Of course, it is quite possible that the bulk complexes isolated from nuclei are different from the proteins binding in vivo at the PCNA intron.
In Vivo Occupancy of E2F Sites and Function of E2F Complexes-E2F/DP heterodimers can function as transcriptional activators or repressors depending on whether or not they are complexed with retinoblastoma family members (50 -53). The in vivo occupancy of E2F sites can be cell cycle-dependent. It was initially observed that promoters that are repressed by E2F⅐Rb family complexes have the functional E2F sites occupied only when the gene is inactive in G 0 and G 1 phase cells (21,42). These repressing E2F complexes are almost invariably located near the transcription initiation sites of cell cycle-regulated genes (22). However, E2F sites can be occupied continuously in all phases of the cell cycle, as it was seen for example in the human thymidine kinase and CDC6 genes (22,54). In these cases, there may be a switch between a repressing and an activating complex with maintenance of identical protein⅐DNA contacts. Here we show that a third situation is also possible. Prominent E2F complexes at the PCNA intron were observed in vivo only in S and G 2 phases of the cell cycle. This is in apparent contrast to the gel shift data, which show equally strong complexes in all cell cycle phases analyzed. Complex formation in vivo may be regulated by additional contacts with proteins binding to adjacent or more distant sites and thus the in vivo data should be much more relevant. The coincidence of in vivo site occupancy and transcriptional induction of the PCNA gene suggests that in the case of PCNA the E2F complex functions largely as a transcriptional activator.
Possible Role of an Antisense Transcript in PCNA mRNA Regulation-Genomic footprinting analysis of intron 1 indicated the existence of a typical cell cycle-regulated promoter, but oriented to generate a potential antisense transcript. Using a sense DNA probe in Northern blot analysis, we indeed detected the presence of a short RNA (500 -600 bp) transcribed in the opposite direction to the sense PCNA mRNA (Figs. 9 and 10). Based on the small size and the assumption that the antisense RNA is not spliced (RT-PCR data), we believe that this divergent transcript covers part of the first intron, all of the first exon, and up to 100 bp of the upstream nontranslated PCNA sequences. The antisense transcript seems to be expressed ubiquitously at low levels in all normal tissues and in cancer cell lines with higher expression in normal tissues (Fig.  10). No apparent cell cycle regulation was detected (Fig. 9). Although containing several short open reading frames, the antisense RNA does not seem to encode a functional protein.
Taken together, these data point to the hypothesis that the antisense transcript may play a role in post-transcriptional control of PCNA gene expression.
Antisense RNA control has been well documented in prokaryotes (55). There is growing recognition that antisense gene transcription occurs also in eukaryotes and might be implicated in gene regulation by repressing the expression of the complementary sense transcript (56 -63). Recently, Sutterluety et al. (64) proposed that a putative antisense transcript, originating from intron 3 of the murine thymidine kinase (TK) gene, can down-regulate the expression of TK mRNA in mouse fibroblasts. The ratio between sense transcription and antisense transcription changed from higher sense transcription in growing Swiss 3T3 fibroblasts to predominant antisense transcription in serum-deprived cells. The mechanisms of such regulation are still poorly understood, but in most cases, the divergent transcripts seem to accumulate in an anti-parallel manner, high expression of the antisense transcript usually corresponds to low expression of sense mRNA. Indeed, the level of the antisense PCNA transcript is generally higher in normal tissues compared with malignant cells, and the opposite is true for the sense transcript (Fig. 10). However, the antisense RNA does not seem to be cell cycle regulated in an opposite way to its sense counterpart (Figs. 1A and 9A). In spite of this, we can still imagine that the ratio of antisense to sense RNA rather than the amount of the single transcripts plays a major role in regulation. In resting fibroblasts, where the levels of PCNA mRNA are low, sense transcripts could be paired with their complementary antisense transcripts produced at constitutive levels, and originated from the divergent promoter in intron 1. Formation of the hybrid RNA may therefore be a major posttranscriptional control mechanism that maintains PCNA expression around basal levels in G 0 and G 1 . Duplex RNAs are very unstable, being more susceptible to attack by doublestranded RNA-specific nucleases. They are substrates for socalled unwinding-modification enzymes, which create abnormal RNA bases through modification reactions that may impede transport or translation or may facilitate degradation (58,65,66). The spanning of the duplex RNA through the exon-intron junction may also suggest a possible interference with the mechanism of splicing. Upon serum stimulation, sense transcription increases, whereas antisense transcription remains constitutive. Only a small amount of sense transcript is now used in duplex formation, whereas the major part of PCNA mRNA may now be efficiently processed. A functional analysis of effects of the antisense RNA on PCNA expression has not yet been done. This may prove difficult using transfection assays because the endogenous antisense RNA cannot be eliminated and, unlike TK Ϫ cells, no cell lines are available that lack the PCNA gene. However, our model is in agreement with results showing that PCNA mRNA stability increases 6 -8-fold after serum stimulation (11) and that removal of intron 1 increases PCNA expression in serum-deprived cells (20). The PCNA antisense RNA is predominantly nuclear (Fig. 9B), which would support the antisense model.
Binding of transcription factors may act as a major event that allows the system to switch from negative to positive regulation. An appealing idea is that the E2F complex, binding at position ϩ583 in a cell cycle-dependent manner, may play a key role in regulating the sense-antisense RNA ratio (Fig. 11). It is evident that the E2F complex does not regulate antisense transcription during cell cycle progression (Fig. 9A). Rather it functions as an enhancer to activate the 5Ј PCNA promoter, perhaps by creating the conformational changes necessary for increased rates of transcription initiation. The new messenger RNA can now be released from the negative control by the antisense transcript and normally processed and exported to the cytoplasm.