Distant Enhancers Stimulate the Albumin Promoter through Complex Proximal Binding Sites*

, The albumin- a -fetoprotein locus epitomizes the main features of transcriptional regulation of fetal and adult hepatocyte-specific genes: developmentally regulated promoters and strong distant enhancers. Full enhancer activity required only a proximal albumin-promoter region containing the TATA box, hepatic nuclear factor 1 (HNF1), and nuclear factor Y (NF-Y) sites. Deletion of the HNF1 site abrogated enhancer and promoter activity, whereas methylation of the site reduced all activity by about 3-fold. Deletion of the NF-Y site attenuated activity by about half, but much of the activity could be replaced by juxtaposition of an upstream region (desig-nated distal element IV). Gel shift and competition analysis demonstrated that binding of architectural factors overlapped NF-Y binding. Moreover, a mutation that eliminated NF-Y binding but only minimally perturbed the surrounding region did not affect enhancer function. In plasmids with a second promoter, the enhancers simultaneously stimulated both albumin and a -fetopro-tein promoters with minimal competition, but surprisingly some mutations in the albumin promoter attenuated expression from both promoters, whereas another uncoupled their expression. With single promoters, the function of the proximal promoter region was controlled by three parameters in the following hierarchy: HNF1

Serum albumin is probably the most characteristic protein synthesized by the mature liver, accounting for more than 10% of total protein synthesis (1). The albumin gene promoter thus provides a paradigm for regulation of transcription in the terminally differentiated hepatocyte, whereas the promoter of the adjacent ␣-fetoprotein (AFP) 1 gene is a model for gene regulation in liver development. Both genes are transcriptionally activated during liver specification. After birth, the AFP gene is selectively silenced, its transcription declining by 3 orders of magnitude. The albumin gene, however, remains very active (2)(3)(4).
The two proteins are structurally related with an evolutionary divergence of 300 -500 million years (5), but unlike many other genes with a similar degree of relationship, albumin and AFP have remained adjacent, physically located on chromosome 4 in man, 14 in rat, and 5 in mouse (6). In all three species, the albumin gene is 5Ј to the AFP gene, and the promoters are separated by about 30 kbp (7)(8)(9). The unusual conservation of the albumin/AFP locus suggests an arrangement necessary for coordinated regulation.
Three strong enhancers lie between the two genes (10,11). These intergenic enhancers are potentially active even when the AFP gene has been developmentally silenced, as demonstrated by studies that combined the enhancers with the albumin promoter. Such combinations are strongly expressed in both fetal and adult livers of transgenic mice (11), as well as cell lines with an adult hepatocyte phenotype that do not express AFP (10). An additional enhancer that lies 10 kbp upstream of the albumin gene promoter (12)(13)(14) regulates the albumin gene. This latter enhancer, in combination with the albumin promoter, shows liver specific expression in transgenic mice (13) and highly differentiated cell lines derived from SV40-transformed hepatocytes (15) but is not active in other cell lines that nevertheless have strong albumin expression (16,17). Therefore, additional, undefined albumin gene transcription controls must account for this expression. The albumin-AFP intergenic enhancers are a strong possibility, because only the AFP promoter is developmentally silenced.
Our analysis of enhancer-promoter interactions in the AFP gene has demonstrated that a discrete promoter-coupling element (PCE) is required for interaction with all three intergenic enhancers (18). This element binds transcription factors FTF/ LRH1 (19) and Nkx2.8 (20,21). We also established a model that combined the enhancers with both albumin and AFP promoters on single plasmids. Surprisingly, both promoters were simultaneously driven at full activity, although a control plasmid with two albumin promoters showed the expected promoter competition (22). Such behavior is not observed with plasmid models of the ␤-globin locus, where promoters compete for a common enhancer. These models combine normally distant transcription regulatory modules and so deal with only a limited aspect of long distance regulation in the ␤-globin locus (23,24). Nevertheless, the plasmid models establish two important properties of the promoters. When combined into the same plasmid, the transcription from each promoter is regulated by simple competition, and the strength of each promoter in the competition is developmentally regulated (25,26).
The albumin-AFP locus represents a distinctive model of long distance gene regulation that has significantly different properties from the ␤-globin system. These properties reflect the specific transcription factors that regulate liver development and the mature hepatocyte phenotype as well as general long distance mechanisms.
The lack of competition implies that the mechanism of interaction with the same enhancers differs fundamentally between the two promoters. Because the albumin promoter lacks the PCE of the AFP promoter, it presumably has other elements that mediate enhancer interactions. Although extensively studied as a discrete transcriptional regulator (27)(28)(29)(30)(31)(32), there has been no previous analysis of albumin-promoter interactions with enhancers. Hence, the rationale of the present study is to define the specific transcriptional regulatory elements in the albumin promoter that mediate enhancer interaction and to demonstrate how these affect the relationship of the two promoters. The studies were carried out in HepG2 cells, which have a phenotype like the fetal hepatocyte and express both genes at high levels.

EXPERIMENTAL PROCEDURES
Cloning-Plasmid pAlb(Ϫ175)CAT was constructed from a previously described albumin-promoter construct (10) by combining a segment from the albumin gene promoter from Ϫ175 (an AluI site, blunt, from a partial digest) to ϩ18 (a BglII site, a synthetic linker site replacing a HindIII site in the CAT gene 5Ј-untranslated region) with the CAT gene as a BglII-BamHI segment; these were joined and cloned in pBluescriptII-KSϩ (Stratagene, La Jolla, CA) at SmaI (blunt) and BamHI linker sites. pAlb(Ϫ84)CAT was cloned similarly, but with a promoter segment from Ϫ84 to ϩ18 (NlaIV, blunt, to BglII). Plasmid pAlb123CAT was constructed by linking three synthetic oligonucleotides (PstI-NdeI, NdeI-AgeI, and AgeI-BglII) and substituting into pAlb(Ϫ175)CAT between a PstI site in the pBluescript linker (Ϫ180) and the BglII site at ϩ18. All site modifications of pAlb123 were constructed by oligonucleotide substitution between specific restriction sites or by double digestion followed by blunting with Klenow DNA polymerase I and blunt ligation. The AFP gene enhancer region was substituted into albumin promoter constructs by joining a 9.7-kbp PstI-BstEI segment of plasmid pBS-AFPCAT (22), which contains the AFP gene enhancer region from Ϫ880 to Ϫ6100 as well as vector and CAT gene sequences, to PstI-BstEI segments containing albumin promoter constructs. These constructs placed the nearest enhancer 1.5-1.6 kbp from the transcription initiation site of the albumin-promoter in various plasmids. Representative plasmids are shown in Fig. 1.
To combine the SV40 early enhancer with various albumin promoters, the enhancer segment was excised from pGL2-Control (Promega, Madison, WI) as a 401-bp BamHI to MfeI segment and inserted into albumin promoter plasmids at the same restriction sites, replacing a 144-bp segment. This placed the enhancer downstream of the CAT gene, 1656 bp from the transcription initiation site (see Fig. 8).
pAlb123RLuc was constructed by substituting a 2154-bp BglII to StyI segment of pGL2-Basic (Promega) into the same sites of pAlb123RCAT. To construct pAFPLuc, the albumin promoter was replaced with an ApaI to BglII (Ϫ995 to ϩ4 of the AFP promoter) segment of pAFP6000, a plasmid that contains the entire AFP enhancer and promoter region fused to the CAT gene in the vector pBluescript KSϩ (Stratagene) (33). The enhancers were then added by substituting a 9016-bp PstI (Ϫ624) to PflMI segment from pAFP6000, which added a ϳ6-kbp enhancer region and substituted the entire vector to produce pE-AFPLuc. To construct dual expression plasmids, pE-AFPLuc was modified by substituting a synthetic oligonucleotide for a 69-bp NheI to KpnI segment in the linker region downstream of the luciferase gene, to add SalI and EagI sites. Various albumin-promoter CAT genes (e.g. 1894 bp for Alb123CAT) were excised with XhoI and EagI and cloned into these sites. This placed the closest AFP enhancer 2079 bp from the AFP promoter and 5161 from the albumin promoter (see Fig. 9).
Cell Transfection-Plasmid DNA was propagated in the methylation-positive Escherichia coli strain DH5␣ and the methylation negative strain JM110. All plasmid DNA preparations were purified through two successive ethidium-bromide-CsCl density gradients. Ca 3 (PO 4 ) 2 transfection of HepG2 cells was carried out as described previously (10). Individual transfection experiments always consisted of a series of identical plates transfected simultaneously under identical conditions. Each determination was the average of two transfections. pSV2CAT was included as a positive control in each experiment, and results are expressed as CAT activity/molar plasmid concentration normalized to the values for pSV2CAT. Luciferase assays were carried out from the same extracts, using the Promega Luciferase Assay System. Plasmid pGL2 control was included as a positive control in each experiment, and the results are expressed as luciferase activity/molar plasmid concentration normalized to the values for pGL2 control.
Gel Shift Analysis-Cell extract and gel shift procedures have been previously described (20). All double-stranded oligonucleotides had TCGA protrusions for labeling by fill-in with Klenow DNA polymerase I. Gel shifts were carried out in the presence of 500 g/ml of poly(dI-dC), unless otherwise indicated. All competition assays contained a 100-fold molar excess of unlabeled oligonucleotide. For supershifts, an antibody to the A subunit of NF-Y was obtained from Pharmingen (San Diego, CA).

Albumin Gene Promoter
Models-Previous studies have demonstrated that the AFP gene enhancers are strongly active when combined with the albumin gene promoter (10,11,22). Our studies established enhancer stimulation of the promoter from distances of 1.7 kbp or greater, using a rat albumin promoter segment extending from ϩ1 to Ϫ308 bp (22).
Comparison of additional plasmid constructs with the same enhancer-promoter distances demonstrated that enhancers combined with promoter segments extending to Ϫ394, Ϫ308, or Ϫ175 (data not illustrated) had essentially the same activity, whereas the combination with a Ϫ84 promoter segment was significantly attenuated (see below). This is consistent with the results of Herbomel et al. (29), who reported that the rat albumin promoter consisted of "six positive regulatory elements concentrated within 130 base pairs" (to position Ϫ153 as numbered in this paper). A slightly longer region (to about Ϫ178) is very highly conserved among mammals (Fig. 1A). To study the function of individual sites in enhancer-promoter interactions, we designed a synthetic promoter ("Alb123") in which presumably silent mutations were introduced to create unique restriction enzyme sites between the characterized transcription factor binding sites (Fig. 1, A and B). Tronche et al. (30) have reported that the GATC sequence from Ϫ53, within the HNF1 binding site, is the target for E. coli methylation of adenosine and that this methylation attenuates HNF1 binding. The studies presented below verified this observation. For full gene activity, it was necessary to propagate DNA in the methylation-deficient E. coli strain JM110. This bacterial strain, however, gave poor yields of relatively lower quality DNA, an effect apparent from the relatively large data ranges (standard deviation indicated by error bars in the figures). For later experiments, a methylation-resistant promoter (Alb123R) was prepared by substituting an A for a G at Ϫ53, as in the human and bovine albumin promoters. The activities of the native Ϫ175, Alb123, and Alb123R promoters were indistinguishable alone or in combination with enhancers (Fig. 1C).
Deletion Analysis-To test the functions of individual promoter sites, specific deletions were constructed and analyzed in transient transfection assays. Fig. 2 summarizes results from several series of experiments. First a new series of sequential deletions (⌬1-⌬4) were compared with a deletion to Ϫ84 of the wild type promoter. Two additional plasmids, ⌬5 and ⌬6, contained discrete deletions of the NF-Y and HNF1 regions, re-spectively. We compared plasmids with and without adenine methylation at Ϫ52. Because methylation caused nearly a 3-fold reduction of gene expression, it provides a major perturbation of the HNF1 site.
The deletions affected enhancer and promoter region transcriptional activity in different ways. Deletions through ⌬3 removed upstream regions that had transcriptional stimulatory activity but did not alter transcriptional stimulation by distant enhancers. Moreover, the strength of enhancer activity was independent of the strength of various promoters, especially because some of the deleted promoters (Ϫ84, ⌬4A) had transcriptional activity indistinguishable from background, yet showed strong stimulation by the distant enhancers. The results suggest a functional division into proximal (TATA, HNF1, and NF-Y sites) and upstream region (C/EBP, DBP, NF1, and distal element III sites). The latter contributes a separate, additive, transcriptional stimulation and functions like an independent enhancer (10, 18).
The proximal region had low transcriptional stimulatory activity, but contained essential components for stimulation by either the distant enhancers or the upstream region. The HNF1 site was most important. Methylation of the HNF1 site reduced activity of the intact promoter by 60% and had a comparable effect on all of the deleted promoters, with and without enhancers, except those from which the HNF1 site had been removed. Moreover, deletion of the HNF1 site removed almost all promoter activity, with and without enhancers. It is important to note that the site itself contributed little direct transcription stimulation (e.g. the ⌬4A plasmids, discussed below).
The region around the NF-Y site also contributed to enhancer promoter interactions. The contribution was less than the HNF1 site, because complete removal caused loss of a little more than half of the total activity of enhancer-promoter combinations. Like the HNF1 site, the NF-Y region contributed little direct promoter stimulation. This is particularly clear The mouse albumin promoter sequences from rat (66), mouse (28), human (67), and cow (68) are aligned along with the synthetic promoter Alb123. Bases that are perfectly conserved among all species are shaded. The approximate locations of known transcription factor binding sites and footprints are shown above the alignments, and the restriction enzyme sites constructed into the Alb123 promoter are listed below the alignments. An E. coli dam methylation site is also marked, at Ϫ53. B, promoter and enhancer plasmid constructs. The plasmid constructs contained either natural or modified albumin promoter-CAT fusions alone (4.7-4.8 kbp) or in combination with the 5.2-kbp AFP gene enhancer region (11.1-11.2 kbp) inserted at the PstI site. The latter placed the nearest enhancer 1.5-1.6 kbp from the albumin-promoter transcription initiation site in various plasmids. The prototype plasmids pALB123CAT and pAFPE-ALB123CAT are illustrated. C, comparison of natural and synthetic promoter activities. The plots show the means Ϯ standard deviation for three (promoters) or four (enhancer ϩ promoter plasmids) separate transfection experiments comparing Alb(Ϫ175) and Alb123 plasmids prepared in JM110 or Alb123R plasmids prepared in DH5␣. from comparison of the ⌬3 and ⌬4 promoter-only plasmids, which had virtually the same activity, although only the former contained the NF-Y region.
Despite a general consistency of findings, comparison of the various deletions indicates two significant problems that led to analysis of additional deletion plasmids and of transcription factor binding in the region. First, the activity of the Ϫ84 deletion was considerably lower than that of ⌬3. One possible explanation for this difference is that the actual NF-Y binding site was compromised by the deletion. Milos and Zaret (34) have previously shown by methylation interference that NF-Y contacts the residues at Ϫ85, Ϫ86, Ϫ90, and Ϫ91, all of which are removed by the Ϫ84 deletion. A second related but more complex discrepancy is the difference between activity of the ⌬4 and ⌬5 plasmids. Both deletions remove the NF-Y site, but, surprisingly, the larger deletion had higher transcriptional activity. This difference was tested repeatedly and found to be highly reproducible. Comparison with the Ϫ84 plasmid and analysis of factor binding (below) suggested that a more upstream region was augmenting activity in certain contexts. Therefore, an additional plasmid pair, designated ⌬4A, was constructed with a slightly larger deletion that removed the most upstream segment between the AflIII and PstI sites. This upstream segment was already absent from the Ϫ84 plasmids. Earlier studies (29) did not reveal a function or binding site in this promoter segment, but Fig. 1 demonstrates that the segment is highly conserved among species, consistent with the possibility that it has regulatory function. The region will subsequently be referred to as distal element IV (DEIV). Removal of DEIV lowered the activity with enhancers to match the ⌬5 deletion and also removed almost all transcriptional activity from the isolated promoter. Apparently, the ⌬4 deletion brought DEIV into the proximal region, where it complemented the activity of existing elements. Notably, in ⌬5, the DEIV was present in its normal position and did not have the same effect. Its local function therefore appears to be architectural. Consequently, the region normally around the NF-Y site presumably has a similar architectural role.
One additional series of deletions was constructed to study how enhancer distance affects activity (Fig. 3). The enhancers were moved 1481 bp closer to each promoter. We previously found that placing the enhancers close to attenuated AFP gene promoters reconstituted enhancer function, enabling discrimination of distance-specific effects from attenuation effects (18). In the current series, the enhancer activities of ALB123, ⌬1, ⌬2, and ⌬5 promoters were distance independent, whereas ⌬3, ⌬4, and ⌬6 promoters showed increased transcriptional stimulation when the enhancers were close. Thus, the presence of proximal region affected short distance interactions, perhaps by competing for promoter elements. Interestingly, the attenuated ⌬5 but not the ⌬4 promoter remained distance-independent, indicating that this local deletion significantly altered critical promoter conformation or function that was reconstituted by the juxtaposition of DEIV.
Analysis of Binding in the NF-Y and DEIV Regions-Protein binding to the native NF-Y and DEIV regions was characterized and compared with the altered regions in the Alb123 and Ϫ84 plasmids. Initially, binding of the region around the NF-Y site was compared with a standard NF-Y binding site from the mouse ␣2(I) collagen gene promoter (Fig. 4).
The albumin promoter site demonstrated the characteristic gel NF-Y gel shift, which was considerably stronger than the prototype collagen gene site. NF-Y was apparent as two closely spaced bands, with the lower one stronger than the upper. Supershift analysis demonstrated that both bands contained Plasmids were derived from pALB123CAT, except for the ⌬4A plasmids, which were derived from pALB123RCAT. Experiments included control transfections with pSV2CAT, and results are corrected for the molar concentration of transfected DNA and normalized to values for pSV2CAT. Three comparisons are presented: 1) the activity of plasmids that contain only the promoter region; 2) plasmids with promoters ϩ enhancers; and 3) enhancer activities, calculated by subtracting the first set of values from the second. A, schematic of deletions. B, methylated DNA. DNA was prepared in E. coli strain DH5␣. Sequence analysis indicates that the only dam methylation site affecting known transcription controls is in the promoter HNF1 site. The data represent two separate experiments for each plasmid. C, unmethylated DNA. DNA was prepared in E. coli strain JM110, except for the ⌬4A plasmids, which were prepared in DH5␣. The data represent three or four separate experiments for each comparison, except for ⌬1, ⌬2, and ⌬4A, which were each studied twice. the NF-Y A subunit. Notably, the regions from the Alb123 and Ϫ84 promoters both showed strong NF-Y binding, indistinguishable from the native site, so the altered expression from the Ϫ84 promoter could not be attributed to reduced NF-Y binding. The gel shifts also demonstrated that the sites bound additional proteins (Fig. 4, X). The collagen gene oligonucleotide also showed X-factor binding, but it was much weaker than the strong pattern of bands obtained with the albumin promoter site. Most of the binding activity in this lower region showed clear competition, indicating that the pattern represented specific binding. The lower bands did not supershift with NF-Y antibodies and were not competed by the collagen gene NF-Y site. There were small differences in the patterns associated with the native, Alb123, and Ϫ84 promoters, but these were difficult to resolve.
Further analysis compared binding in the DEIV region with the NF-Y region, in an effort to explain the anomalous activities of the ⌬4 and Ϫ84 plasmids (Fig. 5). The DEIV oligonucleotide gave a strong pattern of binding that had several properties in common with the lower pattern of bands bound by NF-Y region oligonucleotide. The lower region band patterns were in approximately the same positions with both oligonucleotides. The NF-Y oligonucleotide was a less effective competitor of binding by DEIV, indicating that the latter sites are stronger. Gel shifts of both regions were also studied with reduced poly(dI-dC) concentration. This reduction increased binding but did not otherwise alter the patterns (data not shown). The DEIV bands were stronger but appeared to be specific because of an appropriate self-competition. Moreover, DEIV competed the lower bands from the NF-Y gel shifts without affecting the NF-Y-specific band.
Oligonucleotides that spanned the ⌬4 and Ϫ84 deletions were compared with the wild type regions. These showed no significant new binding, ruling out the possibilities that 1) the NF-Y site was not removed by the ⌬4 deletion or 2) the Ϫ84 or ⌬4 deletions created new binding sites. Taken together, these gel shifts indicated that other than NF-Y, similar but not identical protein complexes bound to the NF-Y and DEIV oligonucleotides. The differences in pattern probably indicated that multiple factors were involved, although most were common to both sites. The transfection analysis suggested that this common binding might function in synergy with NF-Y binding, causing the ⌬4 plasmids to have higher than expected activity, because DEIV region binding replaced activity removed from the NF-Y region. However, the reasons for the low activity of the Ϫ84 deletion plasmids remained unclear.
An additional study characterized differences in binding between the Ϫ84 and wild type NF-Y regions and between the NF-Y and DEIV regions (Fig. 6). This analysis was extended to the region between the C/EBP-DBP and NF-Y sites, which ruled out that a previously undetected activity was removed by the Ϫ84 deletion. However, in competition assays (Fig. 6B), the Ϫ84 NF-Y site was a less effective competitor than the wild type site for bands in the upper part of X but not the lower. This suggested that the binding activities represented distinct sites in different parts of the NF-Y region. A series of oligonucleotides that divided the NF-Y region was then analyzed in similar competition assays (Fig. 6C). These assays discriminated separate binding activities within the complex band pattern. All bands in the region were competed with DEIV and the full NF-Y site. In additional studies (not illustrated), each oligonucleotide was also labeled and directly shifted with and without competitors. Together, the studies demonstrated that the X shifts resulted from separate binding activities at sites that overlap each other and the NF-Y site. The upper band of this region represented binding to 5Ј-side of the NF-Y region, which was altered by the Ϫ84 deletion. The lower bands represented binding by the 3Ј-side of the region. Ϫ84⌬ span the deletions in these two modified promoters. Bases that differ from wild type are shown in lowercase. X denotes the complex groups of non-NF-Y band shifts detected with either DEIV or NF-Y oligonucleotides. Compared with Fig. 4, the nuclear protein concentration has been reduced 2-fold to ϳ3 g/lane to provide better resolution of bands in the X region. Additional studies (not shown) at higher protein concentration did not show additional bands.
The DEIV and NF-Y regions were then analyzed by comparison with two extensive data bases of known transcription factor and other DNA-binding sites (35,36). In simple analysis, the only perfect matches were to NF-Y and HMG-I motifs. A further analysis for less than perfect matches was filtered for motifs that showed partial matches to both the NF-Y and DEIV regions. This identified three additional factors for consideration, TCF (or lymphocyte enhancer factor), YY1, and ANF (Fig. 6A). Oligonucleotides for each were studied in direct and competition binding assays, although only the latter are illustrated (Fig. 6D).
ANF (for albumin negative factor) is a binding activity characterized at two sites in the albumin gene upstream region, at Ϫ8.5 and Ϫ8.6 kbp, reported by Herbst et al. (14). The activity has not been purified or cloned, but the binding sites are distinctive. The stronger site (ANF-A) was chosen for the present analysis; the characteristic band shift (14) comigrated with the upper band of the X region, although other weaker bands were also apparent in our assays. The ALB-NFY and ANF-A oligonucleotides showed similar competition, whereas DEIV competed the band more effectively. The ANF oligonucleotide also selectively competed the upper X region band in ALB-NFY and DEIV gel shifts (not illustrated).
The HMG-I family is comprised of at least three proteins: HMG-I, HMG-Y, and HMG-C. The factors have been reported to bind to a variety of short AT-rich motifs (37), although a more extended bipartite binding site has been reported by Yie et al. (38). For the present analysis, a site from the ␤-interferon enhancer, designated PRD-IV, was chosen. This site also binds an activating transcription factor 2/c-Jun heterodimer. With HepG2 extracts, PRD-IV showed a complex pattern of multiple bands in the position described for HMG-I. Competition with the ALB-NFY site was similar to self-competition, whereas DEIV was a more effective competitor. The PRD-IV oligonucleotide also selectively competed the lower X region bands in ALB-NFY and DEIV gel shifts (not illustrated).
The TCF oligo showed characteristic but weak binding to an upper band (39) using HepG2 extracts, which was not competed by ALB-NFY or DEIV. The YY1 oligo demonstrated a strong characteristic three band gel shift (40), but this shift was neither observed with nor competed by ALB-NFY and DEIV. The complex binding patterns suggest a variety of interacting factors that share some properties with ANF and HMG-I. The former factor has never been definitively characterized, whereas the latter represents several activities, none of which has been characterized in hepatocyte-derived cells. HMG-I is known to have an architectural function, bind AT-rich motifs in cooperation with other binding activities, and form a specific complex with NF-Y (41), so the association with the albumin NF-Y site is appropriate. It is useful to designate the upper X region gel shift band as "ANF-like" and the lower group of bands as "HMG-I-like" pending a more definitive characterization of the activities that bind the NF-Y and DEIV regions. The ANF-like activity binds to the 5Ј-side of the NF-Y region and is reduced by the Ϫ84 deletion. The HMG-I-like activity is primarily localized to the middle and 3Ј-side of the NF-Y region.
Specific Mutations in the NF-Y Region-To discriminate the roles of different binding activities in the NF-Y region, new discrete mutants were constructed in the intact Alb123R promoter. The new plasmids were compared in transfection assays, and oligonucleotides containing the mutations were compared by gel shift (Fig. 7).
To modify the ANF-like binding, m1 replicated the mutated sequence of the Ϫ84 promoter, 5Ј to the NF-Y site. To block NF-Y binding without altering AT-rich motifs, m2 substituted two Gs for the Cs in the core CCAAT motif. To modify HMG-I-like binding without eliminating NF-Y binding, m3 replaced A and T with C and G, bases found at these positions in other NF-Y sites (42). Gel shifts demonstrated that the mutations were largely successful in producing the desired alterations: m1 selectively weakened ANF-like binding; m2 virtually eliminated NF-Y binding; and m3 reduced both HMG-I-like and NF-Y binding. The mutations essentially confirmed that the region contains three overlapping binding activities. However, in transfection assays, the mutations had surprisingly small effects, especially compared with the ⌬5 deletion that completely removed the NF-Y region while keeping the rest of the promoter intact (Fig. 2). m1 had no effect, whereas m2 and m3 each reduced total (enhancer ϩ promoter) activity by only 10%, differences that may not be significant. Moreover, the reduction was accounted for by reduced promoter activity, which was normal for m1, but reduced by 40 and 47% for m2 and m3, respectively. In contrast, ⌬5 reduced both enhancer and promoter activity by more than 50%. Thus, for stimulation of the intact promoter by distant enhancers, the region around the NF-Y site was critical, but the specific contribution of NF-Y was small. Presumably, cooperative interaction of factors that bound the rest of the promoter and near upstream region compensated for the mutations that alter the binding of single components. In contrast, the ⌬5 deletion removed three binding activities and also disrupted local architecture.
Promoter Function with a Heterologous Enhancer-To determine whether the promoter-enhancer interactions were specific for the AFP enhancers, the Alb123, ⌬3, ⌬5, and m2 promoters were combined with the SV40 early enhancer (Fig. 8). This enhancer is active with many promoters but contains no described transcription factor binding sites that are common to the AFP enhancers (43,44). The SV40 enhancer strongly stimulated the intact Alb123 promoter, and the mutated promoters behaved in the same way as with the AFP enhancers. The ⌬3 mutation that removed three upstream sites did not alter enhancer stimulation. At the NF-Y site, the ⌬5 deletion reduced activity by more than half, but mutation m2 that selectively removed NF-Y binding did not affect enhancer stimulation. Finally, HNF1 site methylation significantly attenuated enhancer-stimulated expression through the Alb123, ⌬3, and ⌬5 promoters. Like the AFP enhancers, stimulation by the SV40 enhancer did not require the three upstream binding sites or NF-Y but was critically dependent on HNF1 and on the other activities that bind at the "NF-Y site." Analysis of the Albumin Promoter in a Dual Promoter Model-A final series of studies was carried out in a new version of a dual promoter model that we reported previously (22). In that model, the AFP enhancers were found simultaneously to stimulate both the AFP and albumin promoters at full activity. However, the evaluation was carried out only in DNA-methylated plasmids, in which the albumin promoter had reduced activity. For the current analysis (Fig. 9), an AFP-luciferase reporter system was established in which the enhancers stimulated the AFP-promoter about 40-fold to produce a total activity 8.6-fold greater than the pGL2 control. The latter plasmid, comparable with SV2CAT, was used to standardize the luciferase assays. The AFP luciferase and albumin CAT genes were then combined with the enhancers in dual promoter plasmids. When the Alb123 and AFP promoters were combined, the albumin promoter showed full activity, whereas the AFP promoter showed strong activity close to full activity. There appeared to be a moderate reduction of AFP luciferase compared with the single-promoter plasmid, but this difference was less than the variability of the luciferase assay. The AFP promoter activity was independent of Alb123 promoter attenuation by methylation, which would be expected to increase AFP promoter activity if competition between the two promoters was present. The new Alb123 dual promoter constructs essentially reproduced the findings of our earlier study, demonstrating that the AFP enhancers mediate exceptionally strong activity simultaneously through two promoters with little or no competition.
When dual promoter analysis was extended to mutated promoters, the results were surprising and complex. Because these results were unexpected, each individual plasmid preparation was verified by restriction enzyme mapping and DNA sequencing and by repeat experiments with separately prepared preparations of plasmid DNA: 1) In contrast to the single promoter plasmids, ⌬3 had greatly reduced enhancer-driven expression through the albumin promoter. Remarkably, this albumin promoter mutation also reduced activity through the AFP promoter. 2) On the other hand, the NF-Y site deletion, ⌬5, showed simple behavior, attenuating the albumin promoter to about the same degree as in single-promoter plasmids without affecting the AFP promoter.
3) The NF-Y specific mutation m2 also had significantly reduced albumin promoter expression, unlike its behavior in the single promoter construct. This mutation also attenuated expression from the AFP promoter. 4) Finally, methylation of the HNF1 site selectively weakened the mutated or deleted promoters much more than the unmutated Alb123 promoter. These complex results suggest that the full dual promoter system behaves as more than a combination of enhancer-promoter interactions and competing promoters. Rather, promoter-promoter interactions also appear necessary to explain the observation that albumin promoter mutations can attenuate enhancer-stimulated expression from both promoters. These promoter-promoter interactions include two separate functions, one through a site removed by ⌬3 and the other through NF-Y. DISCUSSION Binding Sites and Sequence Conservation-The defined binding sites (NF-Y, HNF1, and TATA) in the proximal region are highly conserved across mammalian species, even though considerable variation is tolerated in transcription factor binding sites. Moreover, the alignments (Fig. 2) demonstrate constraints on sequence that are not explained by simple binding of the known factors. The following discussion focuses on proximal region sites, although comparison of the more upstream sequences demonstrates similar features.
Even relatively unusual features of the albumin promoter sites are conserved. For example, NF-Y (or CAAT protein 1/CAAT-binding factor) sites center on a consensus CCAAT motif, but the albumin gene site, AACCAATGAAATG, is otherwise atypical. Of 11 sites compiled by Maity and De Crombrugghe (42), the albumin site has the highest A ϩ T content, a consequence of the two underlined bases, which are present in 0 and 1 of 10 nonalbumin sites, respectively.
The HNF1 site also contains conserved atypical features. This factor binds as a dimer to an inverted dyad site with an ideal consensus sequence GTTAAT N ATTAAC (compiled in Ref. 43), although in actual gene sites, the dyads are less than perfect. The albumin promoter site is GTTAATGATCTAC, and the underlined deviating bases are completely conserved, although many other substitutions would be tolerated in these and other positions. An additional 7-bp region on the 3Ј-side of the site is also perfectly conserved, suggesting an overlapping binding site.
The albumin TATA box is also atypical. The usual consensus is TATAWAW (compiled in Ref. 45), whereas the albumin promoter site is TATATTA. Although perfectly conserved, the underlined base is rarely present in TATA boxes. Additional bases are conserved both 5Ј and 3Ј of the TATA box.
In all of these cases, the totally conserved sequences include bases that deviate from the usual binding sites and are not required for the binding of the characterized transcription factors. There are several possible explanations, which are not mutually exclusive: 1) The albumin gene is transcribed at very high levels, and its promoter might contain unusually strong variants of binding sites. However, many single base changes are likely that would allow equally strong sites for these factors. 2) The conserved regions are determinants of promoter architecture that define critical distances and alignment of bound factors. This is not likely to account for most conserved base pairs, because single base substitutions would not significantly alter distances, and many would retain critical determinants of helical structure. 3) Additional binding sites overlap the known binding sites. Binding sites can vary, but the presence of two sites would provide dual evolutionary constraints on bases in the overlap. The need to bind two different factors could also force the occurrence of relatively unusual bases in these sites.
The conserved sequences in the albumin promoter indicate the importance of overlapping sites that have not yet been defined. Some of the overlapping factors probably have an architectural function rather than acting as direct transcriptional activators. Simultaneous binding might occur, but some factors might instead bind sequentially during gene activation. Other binding combinations might represent alternate binding at different developmental stages or negative interactions that down-regulate the gene in nonexpressing cells.
Architectural Factors-The complex effects of deletion at the NF-Y site and juxtaposition of DEIV suggest a "context-dependent" transcriptional function similar to the architectural factor, lymphocyte enhancer factor (46). Because these positional effects may be inconsistent in plasmid models, the anal- ysis was extended to include direct analysis of protein binding. The analysis demonstrated that binding detected in the region of the NF-Y site, other than by NF-Y itself, is important for optimum function of the albumin promoter. Although a complex pattern of several bands was observed, the binding showed features that indicate specificity. Competition was demonstrated, and different oligonucleotides had varying strength as competitors. In addition, longer oligonucleotides from the NF-Y region showed stronger binding of the complexes, suggesting cooperative binding of multiple activities over an extended region larger than a single binding site. Partial purification of HepG2 nuclear extracts (not illustrated) demonstrated that a fraction eluting from heparin-Sepharose at 600 mM KCl produced the entire pattern of band shifts. Because some nonspecific gel shift activity eluted at lower salt, the high salt fraction showed improved specific competition. Together, these observations indicate that the band shifts appear to represent specific binding with relatively high affinity.
The identification is preliminary, but there are several reasons for postulating that some of the binding represents HMG-I family members. First, the high A ϩ T regions conform to known binding sites for HMG-I (37,45). Second, PRDIV, a known binding site for HMG-I from the ␤-interferon gene, produced a similar pattern of bands appropriate for HMG-I (47), and these showed reciprocal competition with the albumin promoter sites. Interspersed HMG-I binding sites have been shown to be an important component of the ␤-interferon gene "enhanceosome" (41,48). The HMG-I factors have an architectural function, binding in the narrow minor groove of AT-rich regions and bending the DNA, which facilitates the binding of transcriptional activators at interspersed sites. Because of the different molecular forms (HMG-I, -Y, and -C, with and without acetylation), multiple bands are expected in gel shifts of native nuclear extracts, and most studies have instead defined the binding patterns of purified or recombinant proteins. Notably, a specific synergistic binding between NF-Y and HMG-I(Y) has been recently described at the CCAAT-box of the ␣2(I) collagen promoter (49,50).
HMG-I does not account for all of the bands in the complex X region gel shift pattern, because the strong top band of the pattern has different competition properties from the lower bands. This binding resembles ANF (14), an activity that has not been correlated with a specific factor. Because observations suggest that the ANF-like and HMG-I-like activities cooperate in binding, it is likely that both represent architectural factors. Such factors might be relatively ubiquitous and show only weak binding individually. An architectural role is evidenced by the complex behavior of DEIV in the albumin promoter deletions. DEIV can replace most of the activity of the NF-Y region when it is juxtaposed to the HNF-I site (compare the ⌬4 plasmids to the ⌬4A plasmids which lack DEIV) but not when it is in its normal position (the ⌬5 plasmids). The effect is thus positional and local, not as expected for a typical transcriptional activator bound at the site. The studies of the ␤-interferon enhanceosome provide a precedent for the importance of similar architectural factors in other promoters and enhancers.
Minimal Promoter Necessary for Enhancer Interactions-The studies in this paper have utilized enhancer-promoter interactions as a model system to define important promoter functions. The albumin promoter alone is strong enough to allow easy analysis of transcription, and previous studies have analyzed its function only as an isolated transcription control region. The present deletion studies indicate that a limited proximal region is sufficient for very strong activation by the FIG. 9. Analysis of dual promoter plasmids. A, structure of dual promoter plasmids, in which the AFP gene enhancers were 2.1 kbp upstream from the AFPluciferase and 5.2 kbp from the Alb123CAT promoter. B, CAT and luciferase assay data. Plasmid DNA was propagated in both methylated and unmethylated forms, as described above. CAT values represent the means Ϯ S.D. of two to three separate experiments, and luciferase data are averaged from five separate experiments. The CAT and luciferase activities are shown on different scales, which have been normalized so that the full values for pE-Alb123CAT and pE-AF-PLuc are equivalent. distant enhancers, but interactions that are more complex are demonstrated by the dual promoter model. The critical proximal region contains known binding sites for three major transcriptional regulators NF-Y, HNF1, and TFIID. Additional binding of architectural and perhaps other factors is also present. A more upstream region, which contains C/EBP, DBP, NFI, and other binding sites, is not required for distant enhancer stimulation in HepG2 cells but does contain at least one activity necessary for noncompetitive sharing of the enhancers with the AFP promoter. Modifications of the proximal region strongly affect stimulation by this upstream region and by distant enhancers. This suggests that the upstream region is an independent local enhancer that depends on function of the proximal region. This local enhancer makes a considerable contribution to albumin gene expression, equivalent to ϳ40% of the SV40 early enhancer plasmid SV2CAT, and the distant enhancers increase this activity to ϳ240% of SV2CAT. This amounts to 6-fold stimulation, but we have previously shown that in strong enhancer-driven systems, the enhancer contribution is described better as an additive relationship (10). In this paper, comparable enhancer activity stimulated the intact albumin promoter 2.4-fold, Ϫ84 promoter 32-fold, the ⌬4A promoter 48-fold (Fig. 2C), and the AFP promoter 39-fold (Fig. 9B). The upstream promoter region of the albumin gene is best considered as an independent enhancer because it adds a strong contribution to transcription that is not required for strong promoter function with other enhancers.
In addition to the TATA box, HNF1 and NF-Y region functions are critical for stimulation by enhancers. HNF1 is clearly the most important. Change of a single base (A to 5-methyl A) reduces activity of all plasmids about 3-fold, an effect greater than deletion of any other promoter region site, whereas deletion of the HNF1 site essentially eliminates both promoter and enhancer function. The site binds homo-and heterodimers of HNF1␣ and ␤, and HepG2 cells contain high levels of HNF1␣ (51). HNF1 is known to be an important regulator of liverspecific gene expression (reviewed in Ref. 52) and to function synergistically with C/EBP in regulation of the albumin promoter (32). The present studies confirm this important function for HNF1 but, surprisingly, suggest that its primary action is indirect, mediated by other factors.
HNF1 may have a general role as a regulator of enhancer function, because it has a similar role in the AFP promoter. In the latter, enhancer stimulation is dependent on two weak HNF1 sites and on the PCE, an element located at Ϫ155 (18). About 80% of enhancer stimulation was eliminated by removal of the PCE, but this stimulation also depended on HNF1, as in the albumin promoter. Inactivation of either AFP-promoter HNF1 site reduced activity by about half, whereas inactivation of both totally inactivated transcription (53)(54)(55). The requirement for HNF1 is virtually identical with the heterologous SV40 enhancer. Moreover, HNF1 is the only obvious component of the proximal-promoter driven system that makes it liver-specific.
The similar behavior of the AFP enhancers and the SV40 enhancer is somewhat surprising because there are no known factors or binding that are common to both sets of enhancers. Only the strongest and most distal AFP enhancer has been well studied. This enhancer is dependent on two liver-enriched factor families, C/EBP and HNF3 (56,57), whereas the SV40 enhancer is regulated by Sp1, NF-B, AP1, Octamer, TEF1, and TEF2 transcription factors (43,44). The AFP enhancers are less well defined than the SV40 enhancer, but neither system is completely resolved, so there may turn out to be critical common regulators or at least common specific interactions with HNF1. However, it seems most likely that HNF1 is essential to the assembly of a preinitiation complex that can interact with the enhancers.
Although secondary to the role of HNF1, the NF-Y region also contributed to activity of distant enhancers. Architecture of the NF-Y region, produced by overlapping binding, was clearly important, because NF-Y itself was not required for the single promoter-enhancer models of this paper. Nevertheless, the high degree of conservation of this atypical site, its strength, its function with the distal promoter region, and its function in dual promoter models all demonstrate the importance of NF-Y. This importance is confirmed by the studies of Milos and Zaret (34), who showed that NF-Y synergistically interacts with the adjacent C/EBP site.
NF-Y is an unusual transcriptional activator, a heterotrimer of ubiquitously expressed A, B, and C subunits (reviewed in Ref. 58). All are necessary for DNA binding. The B and C subunits each contain transcription activation domains that bind TAF II 110 (59). Both A and C have DNA binding domains that contact separate parts of the binding site and also have histone fold motifs that interact with each other to assemble the complex (59 -62). The A subunit also binds P/CAF, a transcriptional coactivator with histone acetyl transferase activity (63), and NF-Y has been shown to "preset chromatin" for histone acetylation (64). Significantly, Currie (41) has demonstrated specific binding between the A-subunit DNA-binding domain and the AT-hook domain of HMG-I(Y).
More Complex Interactions Demonstrated by the Dual Promoter Model-The general goal of this research is to define the specific molecular interactions that regulate the high level gene expression of the albumin-AFP locus, a system that is primarily controlled by strong distant enhancers. It is generally and probably correctly presumed that such systems cannot be fully reconstituted in plasmid transfection models but instead require transgenic animal models. Recognizing this limitation, the current experimental system has been designed around modular plasmids that define and test the individual molecular components of the system and characterize their contributions to function in a manner that reflects in vivo gene regulation. The albumin-AFP locus is very complex, and the transgenic and plasmid models provide complementary sets of information that are both necessary for explaining how these genes function. The dual promoter plasmids, which are designed for modification of individual regulatory sites, demonstrate a highly complex behavior that has many aspects resembling in vivo function. Studies with these dual promoter plasmids suggest that enhancer-enhancer and enhancer-promoter interactions are both part of locus regulation. Based on the observations in this paper, we propose a three-part model as the basis for the next level of experiments to define how the locus works: 1) HNF1 is required for assembly of the preinitiation complex, a function that cannot be replaced by other factors that bind near the promoter or in the enhancers. Weakening the HNF1 site (e.g. by DNA methylation) greatly reduces the level of gene expression but otherwise does not change the overall behavior of the system. 2) Coupling of distant enhancers to the preinitiation complex is independent of NF-Y but significantly affected by local architecture of the proximal promoter region, hence the apparent requirement for specific architectural factors and great sensitivity to local changes in DNA sequence. This sensitivity is demonstrated by marked differences in expression of the ⌬3 and Ϫ84 promoter constructs (Fig. 2, B and  C). The coupling of the distant enhancers to the preinitiation complex might be stabilized by additional binding to elements that are removed from the Ϫ84 promoter or, alternatively, antagonized by the altered architecture of this promoter. 3) Although not required for the distant enhancers in single-promoter plasmids, NF-Y is required for optimum expression of the upstream promoter "enhancer" (a relatively small effect in these experiments) and for full noncompetitive gene expression through two promoters (a much stronger effect). Both effects might result from direct interaction with NF-Y or a specific coactivator recruited by NF-Y, or they might require reorganized chromatin that is dependent on NF-Y (63,64). Most intriguing is the observation that selective loss of NF-Y in the albumin promoter reduces expression through both the albumin and AFP promoters. One possibility is that NF-Y, while bound to the albumin promoter, directly interacts with the AFP promoter. The role of NF-Y might instead be less direct, by modifying how the enhancers interact with the albumin promoter. Studies have suggested cycling of the distant LCR among the ⑀, ␥, and ␤ globin genes (23,24,65), and such cycling may occur in our dual promoter model. In the absence of NF-Y, dissociation of enhancer-promoter complexes during cycling might be slowed, which could reduce activity from both promoters. Both deletion of the distal promoter (⌬3) and selective elimination of NF-Y (m2) simultaneously attenuate both promoters. This suggests an interaction between NF-Y and the distal promoter that is required for full noncompetitive activity through both promoters. The same interaction is probably necessary for activation by the distal promoter enhancer and for the highest level of integrated gene expression stimulated by all of the regulatory components in the two genes. The ⌬5 deletion behaves quite differently, because although it reduces albumin promoter activity, it also uncouples the two promoters so that they show simple competitive behavior. Overall, the two-promoter plasmid models demonstrate an integration of promoter function through direct interactions and that NF-Y mediates an important part of this integration. This suggests intriguing relationships with in vivo gene regulation that can be further defined in the current model system.