Tissue-specific and Hormonally Controlled Alternative Promoters Regulate Aromatase Cytochrome P450 Gene Expression in Human Adipose Tissue*

Estrogen biosynthesis is catalyzed by a microsomal enzyme, aromatase cytochrome P450 (P45Oarom; the product of the CYPl9 gene). The human CYP19 gene comprises nine coding exons, 11-X. Additionally, tissue-specific expression is determined by the use of tissue- specific promoters, which give rise to P45Oarom transcripts with unique 5’-noncoding sequences. In characterize transcripts in by the (rapid amplification cDNA ends) procedure. Four P450arom tran- scripts with unique 5‘ termini were identified, the two unique 5‘-untranslated of the CYP19 gene, 1.3 and sequence

enzyme complex of the endoplasmic reticulum termed aromatase. The aromatase enzyme complex is composed of two polypeptides, NADPH-cytochrome P450 reductase, a ubiquitous flavoprotein of the endoplasmic reticulum, and aromatase cytochrome P450 (P450aroml; the product of the CYP19 gene), a unique form of cytochrome P450, which appears to be present exclusively in estrogen-producing cells (1)(2)(3)(4)(5). The reaction involves three hydroxylation steps and requires NADPH-cytochrome P450 reductase to transfer reducing equivalents to the P450arom, which binds the C19 substrate and catalyzes insertion of oxygen into the molecule, resulting in formation of the C18 estrogen (6)(7)(8)(9)(10).
In most species throughout the vertebrate phylum, estrogen biosynthesis is limited to the gonads and brain. In the human and other primates, however, estrogen production has a wider tissue distribution, which includes adipose tissue (11) and placenta (12). Studies showing an increased fractional conversion of circulating plasma androstenedione to estrone as a function of increased obesity as well as increased age implicate adipose tissue as a significant site of estrogen biosynthesis (13,14) and as the major source of estrogens in postmenopausal woman and in elderly men. Unlike the ovary, estrogen production in adipose is not cyclic but continuous. While the physiological role of extragonadal estrogen production is not clear, there appears to be a relationship between estrogen biosynthesis in adipose and several disease states such as endometrial cancer, breast cancer, and osteoporosis, as well as chronic amenorrhea of obese women and gynecomastia in obese men. Clearly, understanding the regulation of estrogen biosynthesis in adipose tissue will provide insights into the role of estrogens in these conditions. Both cell types found in adipose tissue, namely adipocytes and adipose stromal cells, express aromatase, although much higher aromatase activity and expression of CYP19 transcripts have been observed in stromal cells (15,16). Adipose stromal cells in monolayer culture have been utilized as a model system to examine factors that regulate aromatase in vitro and to study CYP19 gene expression. A variety of factors have been shown to positively affect aromatase activity and P450arom expression in adipose stromal cells, including glucocorticoids as well as CAMP analogs and phorbol esters. The stimulatory effects of dexamethasone require the presence of fetal calf serum (FCS) (17) in the medium while the increases induced by dibutyryl CAMP (Bt2cAMP) or BtzcAMP + phorbol diacetate (PDA) are manifest only in the absence of FCS (18). A variety of growth factors, including epidermal growth factor, fibroblast growth factor, platelet-derived growth factor, transforming P450; FCS, fetal calf serum; kb, kilobase paids); bp, base paids); RACE, The abbreviations used are: P450arom, aromatase cytochrome rapid amplification of cDNA ends; PCR, polymerase chain reaction; RT-PCR, reverse transcriptase-PCR; CAT, chloramphenicol acetyltransferase; PDA, phorbol diacetate.

P450arom Gene Expression in Human Adipose Essue
growth factors a and p, and tumor necrosis factor, have been shown to mimic the inhibitory actions of FCS on Bt2cAMP-or BtzcAMP + PDA-dependent increases in CYP19 expression (18,19). Aromatase regulation by these agents is tissue-specific since in the ovary, glucocorticoids have little or no effect to stimulate aromatase activity2 and phorbol esters inhibit rather than potentiate the Bt2cAMP-induced stimulation of aromatase activity (20). Isolation and characterization of the human CYP19 gene as well as the characterization of tissue-specific alternative promoters has led to the identification of one mechanism to explain the complex and tissue-specific regulation of P450arom expression. The human CYP19 gene spans at least 75 kb and is comprised of nine coding exons, exons 11-X (21-23). Additionally, two noncoding exons, exon 1.1 and 1.2, have been described previously (21). In placenta, the majority of CYP19 transcripts contain exon 1.1 in the 5'-untranslated region (21, 24) and a minor population of transcripts contain exon 1.2 (25). Genomic sequences identified upstream of exon 1.1 have been shown to regulate CYP19 gene transcription in placental-derived choriocarcinoma cells (23, 26). Tissue-specific regulation of CYPl9 expression by alternative promoters was first demonstrated in adipose stromal cells in culture (27). CYP19 transcripts in adipose tissue or in adipose stromal cells in culture do not contain either exon 1.1 or 1.2 sequences; thus promoters upstream of these exons most probably do not serve to regulate P450arom expression in these cells. By primer extension and S1 nuclease analysis using probes spanning the first coding exon, adipose stromal cells maintained in the presence of Bt2cAMP+PDA have been shown to contain transcripts of which some 50% utilize a start site of transcription 26 bp downstream of a TATAlike sequence within the putative promoter-like sequence termed promoter I1 (PII), which is proximal to the start of translation (27). Additionally, the S1 nuclease protection assay was suggestive that at least 50% of adipose CYP19 transcripts may have an alternative transcriptional start site upstream of an exon that is not contiguous with exon 11. The results of PCR experiments suggested that this exon(s) could not be either exon 1.1 or 1.2 and raised the possibility of another untranslated exon in the CYP19 gene, the transcription product of which is spliced onto exon I1 during processing of adipose stromal cell transcripts (27). CYP19 transcripts in the ovary also do not contain the placenta-specific exons 1.1 and 1.2, but the majority of transcripts have a start site of transcription 26 bp downstream of the TATA-like sequence within the putative promoter PI1 (24). Thus alternative promoters regulate CYP19 transcription, giving rise to P450arom mRNAs with tissuespecific 5"untranslated sequences. These untranslated exons are spliced onto a common 3"splice site upstream of the translational start site; thus the coding sequence is the same in all tissues.
The observation that additional transcriptional start sites might be utilized in adipose stromal cells led us to evaluate CYP19 transcripts in these cells as well as in adipose tissue for the presence of novel 5"noncoding sequences. Through construction of cDNA libraries from adipose tissue and adipose stromal cells, exon-specific Northern analysis, and screening of genomic libraries, this work has led to a number of interesting and novel developments including the identification of two unique 5'-untranslated exons of human CYP19, as well as the observation that alternative promoter usage in adipose stromal cells in culture is a function of the hormonal environment under which the cells are maintained. A preliminary account of this work has been presented elsewhere (28).

MATERIALS AND METHODS
Cell Cultures-Subcutaneous adipose tissue was obtained from women at the time of reduction abdominoplasty or reduction mammoplasty. Consent forms and protocols were approved by the Institutional Review Board, University of Texas Southwestern Medical Center at Dallas. Adipose stromal cells were maintained as primary cultures in Waymouth's enriched medium containing Nu Serum (15%, v/v) (Collaborative Research Inc.). Adipose stromal cells were isolated as described (15). Upon reaching confluence the cells were placed in serum-free or FCS-containing (E%, v/v) Waymouth's enriched medium for 24 h. The cells were then treated with dibutyryl CAMP (0.5 mM), dibutyryl CAMP (0.5 mM) + phorbol diacetate (100 nM) in serum-free medium, or dexamethasone (250 nM) in medium containing 15% FCS for 48 h to maximally induce P450arom mRNA levels. Media were removed, and cells frozen at -70 "C until used for RNA isolation. Induction of aromatase activity was verified using the tritiated water release assay (15).
Isolation ofRNA-Total RNA was isolated with minor modifications according to the method of Chirgwin et al. (29) from adipose tissue and adipose stromal cells in culture. Frozen whole tissue was crushed using a mortar and pestle and homogenized in guanidinium thiocyanate, pH 7.0, using a Ultraturax T25 homogenizer. Dishes of frozen confluent adipose stromal cells were scraped in guanidinium thiocyanate. Poly(A)+ RNA was isolated using oligo(dT) affinity chromatography (Type 2 oligo(dT)-cellulose, Collaborative Research).
Rapid Amplification of cDNA Ends-cDNA libraries were constructed from adipose tissue, adipose stromal cells in control serum-free culture medium, medium containing Bt2cAMP+PDA, or medium containing dexamethasone. Construction of RACE cDNAs was performed with minor modifications as described by Frohman et al. (30,31). The first step, first strand synthesis, was performed using 6 pg of total RNA or 1 pg of poly(A)+ RNA, 1 pmol of primer 17 (located in exon 111,

5'-ACTTGCTGATAATGAGTGTT-3'), reverse transcriptase buffer (Life
Technologies, Inc.), 1 mM dithiothreitol, 1 mM dNTPs, 200 units of Moloney murine leukemia virus reverse transcriptase (Life Technologies, Inc.), in a 2 0 4 volume. The primer extension was carried out at 44 "C for 1.5 h. Prior to the tailing step, the single-stranded cDNA was denatured at 65 "C for 5 min, placed on ice, and then tailed at the 3' end with poly(A) using the enzyme terminal transferase. The amplifications were performed using the Perkin-Elmer Cetus buffer system, P450arom primer 24 (located in exon 11, 5'-CTGGTAnGAGGATATGCCC-TCATAAT-37, and Cetus Tag polymerase. An aliquot of the amplified product was run on a 1.8% agarose gel to visually estimate DNA concentrations. The cDNA was then ligated into the pCR2000 vector and transformed into 1NVlaF'-competent cells using the TA Cloning System (Invitrogen Corp.). Positive colonies were screened using the tetramethyl ammonium chloride method as described (32). Positive clones were sequenced using the dsDNA Cycle Sequencing System (Life Technologies, Inc.).
RNA Blot-Hybridization Analysis-Both formaldehyde and glyoxal gel systems were utilized employing the method described by Maniatis et al. (33). Gels were routinely transferred to a charged membrane (Zeta-Probe, Bio-Rad) by capillary transfer as described by the manufacturer. The Northern blots were probed with PCR-amplified and -1abeled 65-75-bp fragments specific to particular exons.
Screening and Sequence Analysis ofan Human Genomic Library"A human genomic library constructed in an EMBL-3 Sp6fl'7 A phage vector (Clontech Laboratories) was plated and screened with minor modifications by plaque hybridization using a PCR-amplified and -1abeled 65-bp fragment specific to exon 1.4 as a probe. Positive restriction fragments were identified by Southern analysis and subcloned into pUC19 (Life Technologies, Inc.). Nucleotide sequence of the genomic clones was obtained by double-stranded sequence analysis employing the dideoxy chain termination method using Sequenase (U. S. Biochemical Corp.).
Amplification of Specific Exons from RNA using Reverse Danscriptase-Polymerase Chain Reaction (RT-PCR)-A single-stranded cDNA was synthesized using 8 pg of total RNA from breast adipose tissue together with 200 units of Moloney murine leukemia virus reverse transcriptase (Life Technologies, Inc.) and 50 pmol of P450aromspecific oligonucleotides corresponding to regions of exon 1.41.3, or 11. The single-stranded cDNAwas then amplified using primers specific for exon 1.4, 1.4 and 11, 1.3, 1.3 and 11, or 11. The amplified products were digested with RNase, separated on a 2% agarose gel, transferred to Zeta-Probe (BioRad), and probed with a radiolabeled oligonucleotide complementary to sequence within the middle of exon 11.

RESULTS
Characterization of cDNA Libraries Prepared from Adipose Tissue and Adipose Stromal Cells-Aputative promoter region, PII, 110 bp upstream of the first coding exon, namely exon 11, was previously identified as a possible regulatory region for CYP19 transcripts in adipose stromal cells in culture treated with Bt,cAMP and PDA (27), and in corpus luteum tissue of the ovary (24). The identification of this putative promoter region was based solely on the identification of a transcriptional start site just downstream of a TATA-like sequence by S1 nuclease protection and primer extension analysis of adipose stromal cell and corpus luteum mRNA. To determine whether promoter I1 was capable of mediating transcriptional regulation in human adipose tissue, 935 bp of PI1 and upstream flanking sequence was ligated to a CAT reporter gene and transfected into human adipose stromal cells. Transfection of the -935AROM-CAT reporter construct into adipose stromal cells resulted in very low levels of CAT activity, which were not appreciably induced by CAMP + PDA, whereas an RSVCAT construct containing the Rous sarcoma virus promoter linked to the CAT gene was capable of initiating transcription in these cells (data not shown). This observation suggested several possibilities: 1) sequences further upstream than -935 are required for promoter I1 activity; 2) sequences downstream of the first coding exon are required for promoter I1 activity; or 3) promoter I1 is not the major promoter regulating CYP19 expression in human adipose.
Based on the evidence of the previously published S1 nuclease protection analysis (27) suggesting different transcriptional start sites in addition to PII-specific CYP19 transcripts, as well as the initial transfection studies employing PII-specific reporter constructs, it appeared likely that adipose stromal cells in culture contain CYP19 transcripts with alternative 5' ends that are regulated by alternative promoters. In order to characterize the genomic regulatory regions upstream of these exons, it was important to identify unique sequences at the 5' ends of CYF'19 transcripts in adipose tissue and in cultured stromal cells. Once these exons were identified, their sequences could be used as probes to determine their location relative to the rest of the CYP19 gene and thus to identify novel regulatory elements upstream of these untranslated exon(s).
Since the possibility existed that adipose tissue transcripts might contain more than one 5"untranslated sequence, it was advantageous to construct the cDNA libraries so that many CYP19 cDNA clones could be readily isolated. Obtaining fulllength cDNA clones of low abundance RNAs has proven to be difficult and laborious using traditional methods of cDNA cloning. However, utilizing a relatively new method of cDNA synthesis by PCR, a large number of cDNA clones can be identified rapidly. The method termed RACE (rapid amplification of cDNA ends) (30, 31) allows for synthesis of primer-extended cDNA libraries using the DNA polymerase chain reaction technique to amplify copies of the sequence between a single region in the transcript and the 3' or 5' end. The minimum information required for this amplification is a single short stretch of sequence within the mRNA to be cloned. However, since the RACE procedure utilizes the PCR method, which can give rise to differential amplification of particular transcripts, no firm conclusions can be reached as to the relative abundance of any newly identified sequence based solely on RACE identification. RACE cDNA libraries were constructed from adipose tissue, as well as adipose stromal cells that were either left untreated or treated with dexamethasone or Bt2cAMP + PDA. By making libraries from cells in culture under the various hormonal conditions, one can then establish if there are new 5' termini and thus determine if alternative promoters are employed in response to changes in the hormonal environment of the cells.
A number of cDNA clones from each library were sequenced in order to estimate the distribution of the alternative 5' sequences. As indicated in Table I, in addition to the promoter 11-specific sequence, four new 5' sequences were identified (Fig.  1). These new sequences are referred to as 1.3 and 1.4 in the order that they were discovered, and 1.3-truncate and I.fl.2. All of the new sequences were spliced onto exon I1 at the same 3"splice junction as exons 1.1 and 1.2, upstream of the start of translation, and thus would be expected to encode the same protein. The distribution of the various 5' sequences seemed to be influenced by the hormonal treatment of the cells (Table I).
In adipose tissue, 1.4, 1.3, and 1.3-truncate sequences are present. Three cDNA libraries were made from adipose tissue; two were from breast adipose of two different patients, and the other from thighkalves. No 1.4-containing CYP19 transcripts were identified in the library made from thighkalf tissue, and no 1.3 truncate-containing CYPl9 transcripts were identified in the libraries made from breast adipose tissue. The difference in distribution of 5' ends in the three adipose tissue libraries could be due to patient to patient variation, or else it could be a function of tissue localization. In the stromal cells in culture, the choice of 5' termini appears to be dependent on the hormonal environment of the cells. Most interestingly, in cells expressing PII-specific sequences, namely cells treated with CAMP + phorbol esters in the absence of serum, no 1.4-specific sequences are detected. Conversely, in cells expressing 1.4-specific sequences such as dexamethasone-treated cells in the presence of serum, no PII-specific sequences are observed. P450arom transcripts containing exon 1.3 are present in adipose tissue as well as in cells in culture under all conditions. Sequence analysis of the cDNAs revealed that the 1.3 and 1.3-truncate cDNAs were identical at their 5' ends ( Fig. 1). Based on the longest cDNA isolated, the 1.3-specific sequence is 205 bp and the 1.3-truncate-specific sequence contains 99 bp of the most 5' sequence of exon 1.3. The possibility exists that both transcripts are driven by a common promoter. Based on the abundance of 1.3 cDNA clones versus 1.3-truncate clones, it is likely that P450arom mRNA species containing 1.3 sequences are present in higher levels than those containing 1.3-truncate

~~~A~n P r o I l e H i s~~A s n l l e T h r S e r h r s e r~l~v~l~r o G l~A l~n~t~~a A~a A~~~h~~~t~~a v a~~~~~~~
LeuThrGlyLeuPheLeuLeuvalrrpAsn'IyrCluGlyrse=se=Ilep=aG Exon 1 . 4 . exon 1 . 2 and Exon 11

a l L e u G l~e t L e u A s n P r a I l e H i s T y r A s n~~~r h r s~=~~~v~l~r~~~A~a~~t p~a A~~A~~~h~~ e t P~~V a l L e u L e u L e u T h r G l y L e u P h e L e~~~~v~l r~A~~~~e~~G~y~~s e~s e~~l~p~~G
Promoter 11-8Decific Sequence Northern Analysis Using Exon-specific Probes-By means of construction and characterization of cDNA libraries, new 5' sequences were identified; the distribution of these sequences appeared to be influenced by the hormonal treatment of the cells. As mentioned previously, characterization of cDNA clones generated by the RACE procedure enables identification of new sequences but may not give an accurate estimate of the relative abundance of transcripts with any particular 5' sequence. In order to obtain better insight into the relative expression of variant 5' ends under different physiological circumstances, Northern analysis was performed using sequences specific to  (Fig. 3). The RNA blot in Fig. 4, which contained 40 pg of poly(A)+ RNMane, was also probed with a PII-specific fragment, and again no hybridization was detected in poly(A)+ RNA from dexamethasone-treated adipose stromal cells (data not shown). This is the first conclusive evidence of differential expression of 5' termini under different culture conditions. While a single cDNA clone amplified from control cells maintained in the absence of FCS contained PII-specific sequence, no hybridization to PII-specific probes was detected by Northern analysis, suggesting that only a very few P450arom transcripts in untreated cells contain this sequence. As can be seen in Fig. 3, Northern analysis using a probe specific for both 1.3 and 1.3-truncate sequences revealed hybridization to RNA from control cells maintained in the absence of FCS, from cells treated with Bt2cAMP, or Bt2cAMP + PDA, but not from glucocorticoidtreated cells. This was unexpected since 1.3-containing cDNA clones were identified in libraries made from RNAisolated from dexamethasone-treated adipose stromal cells (Table I). However, subsequent Northern analysis of 40 pg of poly(A)+ RNA from dexamethasone-treated adipose stromal cells, as well as untreated cells in the absence and presence of FCS, revealed hybridization to an exon 1.3-specific probe, although for dexamethasone-treated samples the level of 1.4-specific transcripts appeared to be severalfold greater than the level of 1.3-specific transcripts (Fig. 4). The fact that the abundance of 1.3-containing transcripts in dexamethasone-treated cells is low compared to the abundance of 1.4-containing transcripts may explain the lack of 1.3-containing transcripts detectable in Fig. 3, where only 20 pg of poly(A)+ RNA was used. Alternatively, since the RNA used in the experiment shown in Fig. 3 was prepared from a pool of breast and abdominal adipose tissue from several subjects, whereas that used in Fig. 4 was from breast adipose tissue from two different donors, there may be a region-specific difference in the relative expression of these transcripts. A 1.4specific probe hybridized only to RNA from dexamethasonetreated cells in the presence of FCS (Figs. 3 and 4). This result is consistent with the absence of P450arom cDNAs containing exon 1.4 in libraries from Bt2cAMP-or Bt2cAMP+PDA-treated cells. Additionally, no hybridization of the 1.4-specific probe was detected in transcripts from untreated cells either in the absence or presence of FCS (Fig. 4). While dexamethasone-dependent increases in aromatase expression occur only in the pres-ence of FCS, the lack of 1.4 transcripts in untreated cells maintained in the presence of FCS suggests that glucocorticoids have a direct action to regulate expression of 1.4-containing transcripts, together with growth factor(s) present in the serum. On the other hand, a probe specific for exon 1.2 failed to hybridize under any conditions, suggesting that I.4A.2 containing transcripts are present in very low abundance (data not shown). This observation is consistent with the low number of cDNA clones that contain this particular 5' end. These results obtained employing exon-specific Northern analysis are summarized in Table 11.

P450arom Gene
Northern analysis showed that 40 pg of poly(A)+ RNA from adipose tissue failed to hybridize to any CYP19 probe (Fig. 3,  lane 11, while a probe specific to the adipose-specific glucose transporter, Glut 4 (34, 351, showed readily detectable hybridization (data not shown). Thus although the RNA is intact, CYP19 transcripts in whole adipose tissue are present at levels too low to be detected by Northern analysis. As an additional means to verify the presence of exons 1.4 and 1.3 in adipose tissue, RT-PCR was used to amplify exons 1.3 and 1.4 separately and 1.3 or 1.4 together with exon I1 (Fig. 5). Total RNA from breast adipose tissue was used as a template for first strand cDNA synthesis. Exon 11, the first coding exon, was amplified alone as a positive control. The untranslated exons 1.3 and 1.4 were amplified together with exon I1 and so should represent only spliced products. As shown in Fig. 5 (lanes 4 and 61, both exons 1.3 and 1.4 were readily amplified from adipose tissue in agreement with the observed RACE cDNAs. In lane 6 three bands can be observed. The largest and faintest band, at 488 bp, represents unspliced RNA products containing exon 1.3, a 100-bp intron, and exon 11. The second band, a t 388 bp, represents exon 1.3 and I1 together. The smaller band, a t 282 bp, represents the 1.3-truncate sequence spliced onto exon 11. This experiment is in agreement with the results obtained using the RACE cDNA clones, namely that in whole breast adipose tissue, CYP19 transcripts have alternative 5' ends, which include exons 1.3 and 1.4. However, since both analyses were based on PCR, no conclusions can be made with respect to the relative abundance of these transcripts.

Genomic Localization of New CYPl9 Untranslated Exons
-At this point it was evident that alternative CYP19-specific 5' termini are expressed in a tissue-specific fashion and that, additionally, these alternative 5' sequences are expressed in adipose stromal cells cultured under different conditions. Since promoter sequences are generally located just 5' of the start site of transcription, expression of these alternative 5' ends may be regulated by alternative promoters. Having identified these new 5' sequences, it was then necessary to determine their location in the CW19 gene, in order to identify the putative promoter sequences regulating expression of these alternative CYP19 transcripts.
Upon comparison of exon 1.3 to sequences upstream of exon 11, exon 1.3 was found to be located just upstream of exon I1   (Fig. 6). A 100-bp intron exists between exons 1.3 and I1 such that promoter I1 and some of its upstream regulatory sequences are included as exonic sequences in exon 1.3-containing P450arom transcripts. Based on the largest cDNA clone isolated containing exon 1.3, exon 1.3 appears to be approximately 205 bp. It should be noted that a second proximal TATAA sequence exists in the gene 20 bp upstream of the end of the sequence contained in the longest exon 1.3-containing clone (27) and 325 bp upstream of the common splice junction in exon 11. Furthermore, a sequence, CAAAAT, is present 37 bp upstream of this sequence. Whether or not this region is the promoter sequence regulating expression of 1.3-containing transcripts will be determined by primer extension and S1 nuclease protection analysis, as will the exact size of exon 1.3.
A comparison of exon 1.4 to sequences upstream of exon I1 revealed no homologies. In addition, Southern analysis of genomic fragments containing exons 1.1 or exon 1.2 indicated that exon 1.4 is not located in these regions of the P450arom gene.
Since the genomic clones containing exons 1.1 and 1.2 did not overlap, a gap remained in the CYP19 gene in the region between exon 1.1 and exon 1.2. In consideration of the possibility that exon 1.4 was located in this region, a human genomic library was screened in hopes of isolating the remainder of the gene. An EMBL-3 SP6R7 human genomic library was screened using a PCR-labeled 65-bp exon 1.4-specific fragment as a probe. A single, 16-kb positive clone was obtained after screening 450,000 plaque-forming units (Fig. 7). The 16-kb insert was cut out of A using SacI to release 1.3-, 8.5, 1.5, and 4.5-kb fragments. Southern analysis using the exon 1.4 probe indicated the presence of exon 1.4 in the 4.5-kb SacI genomic fragment.
In an attempt to determine if the 1.4-containing genomic fragment overlapped with genomic clones containing either exons 1.1 or 1.2, or both, Southern analysis was performed using random-primer labeled fragments corresponding to the 3' end of the exon I. 1-containing genomic fragment or the 5' end of the 1.2-containing genomic fragment. No hybridization was detected with the 1.2-containing fragment, indicative that the 1.4 clone did not overlap this region of the gene. However, the 1.1-containing genomic fragment did hybridize to the 1.4-containing genomic clone, verifying that these two clones overlapped. Through restriction mapping and Southern analysis, untranslated exon 1.4 was estimated to be located 20 kb downstream of exon 1.1. The size of the overall gene is therefore estimated to be at least 75 kb. Our current knowledge of the organization of the CYP19 gene upstream from exon I1 is summarized in Fig. 6, which additionally shows the various splicing possibilities that we have detected.
Characterization of Exon Z.4-The 4.5-kb SacI fragment containing exon 1.4 was subcloned and sequenced (Fig. 8). Upstream from the end of the longest sequence found in the cDNA clones there are no obvious TATA, CAAT, or GC-rich sequences, which might serve as possible promoter regions. At the same time, the size of 1.4-containing transcripts, as indicated by Northern analysis (Figs. 3 and 41, is such that we would not expect the exonic sequence to extend much further upstream. Primer extension and S1 nuclease protection analysis will have to be undertaken to define the promoter sequences in this region of the gene. DISCUSSION Tissue-specific Expression of Human P450arom-Aromatase is expressed throughout the entire spectrum of the vertebrate phylum. In most species however, expression is confined to the gonads and the brain. In humans and some other higher primates, expression occurs additionally at other sites, in particular, placenta and adipose tissue. Based on the results of this and previous studies, we conclude that in the human, tissuespecific expression of the CYP19 gene is under the control of tissue-specific promoters. The resulting transcripts are generated by means of alternative splicing. Thus, expression in ovary appears to be under the control of a promoter, which is proximal to the start of translation (241, similar to the situation in rat (36) and chicken (37). This is likely, therefore, to be the primordial promoter regulating CYP19 expression. When in the course of evolution the human placenta developed the ability to synthesize estrogens, this promoter was not utilized. Instead, transcripts from placenta contain sequences that are at least 35 kb upstream from the start of translation, apparently under the regulation of a distal promoter 1.1 (21). This helps to explain the differences in the regulation of aromatase expression in these two tissues. By contrast, transcripts in adipose contain two different 5' termini, which we have named exon 1.  splicing event and, in fact, contain some of the promoter I1 region as exonic sequence. Exon 1.4 is located 20 kb downstream from exon 1.1 and therefore is at least 15 kb upstream from the start of translation. However, since the clone bearing this sequence does not overlap those containing exons 1.2 and 11, the actual distance is not yet known.

t 9 C a t t C t A A C C A T Q A C A Q C C A C A Q T C A Q Q A C A C A A A A A A C A A A Q T Q~C C T
Additionally, a number of other splicing events occur in various tissues giving rise to minor CYP19 transcripts. For example, in placenta, a small percentage of transcripts contain exon 1.2, located 9 kb upstream of the start of translation (25). Additionally, in adipose tissue there is a truncated form of 1.3, which contains 99 bp of the most 5' sequence of exon 1.3, and in adipose stromal cells in culture two clones have been sequenced, which contain exon 1.4 spliced upstream of exon 1.2, consistent with the order in which these sequences are located in the gene. However, these all appear to be rare transcripts resulting from infrequent splicing events, as judged by their failure to be visualized by means of exon-specific Northern analysis, and their physiological significance, if any, is unclear. Once again, it should be emphasized that all of these untranslated exonic sequences are spliced onto a common 3"splice junction upstream of the start of translation. Consequently, the coding region and, hence, the protein product are identical in each of the tissue sites of expression.
Adipose-specific Expression of P450arom-The identification of two previously uncharacterized exonic sequences in the 5'untranslated region of adipose transcripts of CYP19, namely exons 1.3 and 1.4, defines the nature of CYP19 transcripts in this tissue. Interestingly, the apparently different distribution of 1.4 and 1.3 transcripts in breast versus thighlcalf adipose tissue gives rise to the possibility of alternative adipose CYP19 transcripts, the expression of which is dependent on tissue location. Based on the observation of relatively high P450arom expression in lower body fat versus upper body fat,3 one can speculate that the use of alternative promoters based on tissue location may affect P450arom activity.
It is likely that these newly discovered P450arom transcripts arise due to the use of specific promoters, but these have not been characterized. Upstream of promoter I1 there is a second TATA-like sequence and CAT-like sequence (21, 24,27), which may constitute part of promoter 1.3; however, the localization of these elements relative to the transcription start site for exon 1.3 has not been determined as yet by means of either primer extension or S1 nuclease protection analysis. If this is the case, then promoter 1.3 and promoter I1 share most of their 5'-regulatory sequences in common. In spite of this, the expression of transcripts with 5' ends that are specific for each of these promoters in adipose stromal cells in culture is quite different and is dependent upon the hormonal mileau. Similarly, a promoter region responsible for expression of transcripts containing exon 1. 4 has not yet been identified.
Hormonal Regulation of Promoter Selection in Adipose Stromal Cells in Culture-One of the most intriguing aspects of the present study is the observation that the distribution of 5' termini of CYP19 transcripts in adipose stromal cells in culture is a function of the culture conditions. In particular, promoter 11-specific transcripts are present only in cells maintained in the presence of dibutyryl cyclic AMP plus phorbol ester or else dibutyryl cyclic AMP alone, in the absence of FCS. By contrast, transcripts containing exon 1.4 are absent under these culture conditions. On the other hand, exon 1.4-containing transcripts are present in cells maintained in the presence of dexamethasone plus FCS, whereas promoter 11-specific transcripts are entirely absent under these conditions. Exon 1.3-specific transcripts on the other hand, appear to be present under all culture conditions including those in which cells are maintained in the absence of stimulatory factors. Based on these findings, we propose that expression of transcripts containing exon 1.4 is glucocorticoid-specific and, in addition, requires the presence of serum or growth factors. Thus, glucocorticoid stimulation of expression mediated by putative promoter 1.4 would lead to the formation of transcripts containing exon 1.4 in their 5' termini. By contrast, expression of promoter 11-specific sequences appears to be cyclic AMP-mediated. This concept is consistent with the finding that, in the ovary, transcripts specific for promoter I1 are uniquely present (24), whereas those specific for exon 1.4 are ~ndetectable.~ CYP19 expression in human granulosa cells is known to be stimulated by cyclic AMP analogs but not by glucocorticoids. Similarly, a sequence apparently identical to exon 1.4 has recently been reported to be present in 5' termini of CYP19 transcripts in human skin fibroblasts, cells in which aromatase activity is known to be induced by glucocorticoids (38).
An important question that then arises is whether or not enhancement of expression of transcripts from promoters 11 and 1.4 by cyclic AMP and dexamethasone, respectively, is suf-T. M. Price, unpublished observations. C. Jenkins, unpublished observations.

P450arom Gene Expression in
Human Adipose Tissue ficient to determine the splicing pattern required to produce the respective 5'-termini containing promoter 11-specific sequence and exon 1.4, or whether additional factors are required.
Selection of alternative splicing pathways has been found to be an important regulatory step in expression of a number of genes, and regulated splicing can function as an odoff switch in gene expression. The best examples of this to date have been described to occur in Drosophila (39), where several genes controlling sex determination function in a regulatory cascade that operates principally at the level of splicing. Three genes (Srl, n u , and Du2) encode proteins that directly or indirectly modulate splice site selection (40)(41)(42). Recently a protein factor, alternative splicing factor, was isolated from human HeLa cell extracts and shown to alter the selection of alternative 5"splice sites in an SV40 early pre-mRNA when added to in vitro splicing reactions (43)(44)(45)). An important question that therefore arises regarding hormonal selection of expression of specific CW19 transcripts is whether the mechanism of alternative splicing is subject to hormonal regulation distinct from and in addition to that governing selection of promoter utilization. Whatever the mechanisms involved, however, to our knowledge this is the first time that in vitro hormonal regulation of alternative promoter usage has been described in a mammalian gene. Clinical Considerations-Although the physiological role of aromatase activity in human adipose tissue is not clear, it is likely that aromatase activity has a role to play in the maintenance of bone mineralization and, hence, prevention of osteoporosis. However, as discussed earlier, the pathophysiological sequelae are more than apparent, namely the implication of estrogen synthesized in adipose in the etiology of both endometrial and breast cancer (46). We have studied the distribution of CYP19 transcripts in breast fat, in relation to the localization of a tumor, and have provided evidence that expression is higher in regions proximal to the tumor as compared to distal region^.^ This is consistent with previous studies in which similar regional variations in aromatase activity were reported (47). These observations lend credence to the view that there is cross-talk between tumor cells and the surrounding adipose cells in terms of control of local production of estrogens, as we have suggested previously (19). We also have shown recently that P450arom expression in adipose tissue displays striking regional as well as age-dependent variation^.^ As indicated in the present study (Table I), there may be regional variations in promoter usage in different adipose tissue sites, in that transcripts containing exon 1.4-specific sequence appear to be present only in breast fat, whereas 1.3-specific sequences are present in both upper and lower body fat. It remains to be determined whether alternative promoter usage provides a basis for the age-and region-specific distribution of P450arom expression in human adipose tissue.