Characterization of a Novel Type of Human Microsomal 3 a -Hydroxysteroid Dehydrogenase

We report characterization of a novel member of the short chain dehydrogenase/reductase superfamily. The 1513-base pair cDNA encodes a 319-amino acid protein. The corresponding gene spans over 26 kilobase pairs on chromosome 2 and contains five exons. The recombinant protein produced using the baculovirus system is local-ized in the microsomal fraction of Sf9 cells and is an integral membrane protein with cytosolic orientation of its catalytic domain. The enzyme exhibits an oxidoreductase activity toward hydroxysteroids with NAD 1 and NADH as the preferred cofactors. The enzyme is most efficient as a 3 a -hydroxysteroid dehydrogenase, converting 3 a -tetrahydroprogesterone (allopregnanolone) to dihydroprogesterone and 3 a -androstanediol to dihydrotestosterone with similar catalytic efficiency ( V max values of 13–14 nmol/min/mg microsomal protein and K m values of 5–7 m M ). Despite ; 44–47% sequence identity with retinol/ 3 a -hydroxysterol dehydrogenases, the enzyme is not active toward retinols. The corresponding message is abundant in human trachea and is present at lower levels in the spinal cord, bone marrow, brain, heart, colon, testis, placenta, lung, and lymph node. Thus, the new short chain dehydrogenase represents a novel type of microsomal NAD 1 -dependent 3 a -hydroxysteroid dehydrogenase relative to microsomal protein ( m g: m g) in a final reaction volume of 36 m l. The reaction was stopped by the addition of protease inhibitor phenylmethylsulfonyl fluoride to the final concentra- tion of 2 m M . Samples of the reaction mixture were analyzed by Western blotting and by activity measurements using 5 m M androsterone as substrate.

Oxidation and reduction of hydroxyl and ketone groups in position 3 on naturally occurring steroids play an important role in regulation of intracellular levels of biologically active steroid hormones. For example, in gonads, 3␣-hydroxysteroid oxidoreductase activity is responsible for maintaining the balance between a potent androgen, 5␣-dihydrotestosterone, with a ketone group in position 3 and a weak androgen, 3␣-androstanediol, which has a hydroxyl group in the same position ( Fig. 1; reviewed in Ref. 1). In the central nervous system, 3␣-hydroxysteroid oxidoreductase activity controls the amount of a potent neurosteroid, 3␣-tetrahydroprogesterone (also called allopregnanolone), which serves as an allosteric regulator of all ␥-aminobutyric acid type A receptors and potentiates ␥-aminobutyric acid mediated chloride conductance (1). 3␣-Hydroxysteroid oxidoreductase activity has been described in the cytosolic and microsomal fractions of a number of human and animal tissues (2)(3)(4)(5)(6)(7)(8)(9)(10)(11). In addition to the different subcellular localization, cytosolic and microsomal enzymes appear to have a distinctively different cofactor preference. Cytosolic 3␣hydroxysteroid oxidoreductases exhibit a preference for the phosphorylated nucleotides as cofactors (NADP ϩ /NADPH), whereas microsomal enzymes prefer NAD ϩ /NADH (2)(3)(4)(5)(6)(7)(8)(9)(10)(11). Because the predominant forms of nucleotide cofactors in the cells are NAD ϩ and NADPH, it is generally believed that the NAD ϩdependent dehydrogenases function mainly in the oxidative direction, whereas the NADPH-dependent enzymes function as reductases (12). Thus, the balance between the oxidative and reductive activities will determine the local concentrations of bioactive compounds in specific tissues. To date, at least four types of cytosolic 3␣-hydroxysteroid dehydrogenases (HSDs) 1 have been identified in humans (1,13,14). All four cytosolic enzymes, named AKR1C1-AKR1C4, are members of the aldoketoreductase gene superfamily and prefer NADPH as cofactor (13). Liver is the only tissue that contains all four isoforms at similar levels (13). Extrahepatic tissues such as lung, prostate, uterus, mammary gland, brain, small intestine, and testis vary significantly in their composition and relative amounts of individual isoforms (13).
The identity of the enzymes responsible for the NAD ϩ -dependent 3␣-hydroxysteroid oxidoreductase activity found in microsomes has remained elusive until recently, when several newly discovered members of the short chain dehydrogenase/ reductase superfamily were shown to stereospecifically oxidize the 3␣-hydroxyl group on 3␣-androstanediol (15)(16)(17)(18)(19)(20) and allopregnanolone (21,22). In contrast to cytosolic 3␣-HSDs, these membrane-bound short chain dehydrogenases exhibit a strong preference for NAD ϩ as cofactor. Previously, we characterized the properties of two human microsomal short chain dehydrogenases that are capable of oxidizing 3␣-androstanediol to 5␣dihydrotestosterone and allopregnanolone to 5␣-dihydroprogesterone (20,21). The range of potential physiological substrates for these enzymes (RoDH-4 and RoDH-like 3␣-HSD) is not limited to 3␣-hydroxysteroids because they can also oxidize all-trans-retinol and might contribute to the biosynthesis of a potent morphogen, all-trans-retinoic acid (20,21).
Besides RoDH-4 and RoDH-like 3␣-HSD, human cis-retinol * This work was supported by the National Institute on Alcohol Abuse and Alcoholism Grants AA00221 and AA12153 (to N. Y. K.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  dehydrogenase Rdh5 was also shown to oxidize 3␣-hydroxysteroids, although with lower catalytic efficiency than cis-retinols (16). Rdh5 was found to localize on the lumenal side of the endoplasmic reticulum (23) and was implicated in the biosynthesis of 11-cis-retinal (24), the cofactor for vision, and 9-cisretinoic acid, the activating ligand for retinoid X receptors (25). The three human retinol/sterol dehydrogenases share ϳ50% amino acid sequence identity, similar genomic structure, and chromosomal localization on chromosome 12 (26 -28). Individual enzymes exhibit distinctively different tissue distribution patterns. RoDH-4 is primarily expressed in human liver and skin (20,29). RoDH-like 3␣-HSD is present in liver, lung, prostate, testis, spleen (15), spinal cord (21), and various areas of human brain (21). Rdh5 appears to be ubiquitously expressed in a wide variety of tissues: liver, mammary gland, colon, thymus, small intestine, kidney, and others. (16,25). Remarkably, liver contains the highest levels of mRNAs for all three enzymes. Here, we report molecular cloning and characterization of a novel human 3␣-hydroxysteroid dehydrogenase. We show that, in contrast to the previously identified enzymes, this human 3␣-hydroxysteroid dehydrogenase is not active toward retinoids and exhibits different membrane topology as well as different tissue distribution.

EXPERIMENTAL PROCEDURES
Cloning of the Full-length cDNA-cDNA clone W17165 was identified in the expressed sequence tag data base of GenBank TM based on its similarity to human RoDH-4 cDNA. Sequencing of the clone obtained from American Type Culture Collection revealed that it lacked the 5Ј-end (started at nucleotide 346 in Fig. 2). The missing part was obtained by rapid amplification of cDNA ends (RACE) using 5Ј-RACEready human liver cDNA (CLONTECH) as template. The internal genespecific primers (TGACATTCTCTGGGTCGGTCA and TCTTCAC-CCACTGGGCAGTC; both antisense; indicated by arrows in Fig. 2) were designed based on the 5Ј-end sequence of clone W17165. The gene-specific primers were paired with the anchor primer (CLON-TECH), and the cDNA was amplified in two sequential reactions using Taq polymerase (PerkinElmer Life Sciences). The amplifications were performed for 30 cycles as follows: denaturing at 94°C for 1 min, annealing at 60°C for 1 min, and extension at 72°C for 5 min. The ϳ400-base pair cDNA product was subcloned into M13mp18 and sequenced. This cDNA fragment encoded amino acids 1-90 and contained 153 base pairs of the 5Ј-untranslated region (Fig. 2). The complete nucleotide sequence was submitted to GenBank TM with accession number AF343729.
To obtain the full-length cDNA, human liver and heart mRNAs were amplified by PCR using Pfu polymerase (Stratagene Cloning Systems, La Jolla, CA) and primers GGGGGATCCATGCTCTTTTGGGTGCT-AGG (sense primer; the BamHI site is underlined) and TTAGAATTC-TCACACTGCCTTGGGATTAG (antisense primer; the EcoRI site is underlined). Thirty cycles of amplification were performed as follows: denaturing at 94°C for 45 s, annealing at 55°C for 45 s, and extension at 72°C for 6 min. The amplified cDNAs were subcloned into BamHI/ EcoRI restriction sites of pVL1393 transfer vector (PharMingen, San Diego, CA). The transfer vectors were sequenced to verify the cDNA sequence of the gene of interest. The sequences of the cDNAs obtained from liver and heart were identical. The structure of the corresponding gene was determined by searching Human Genome GenBank TM data base using the full-length cDNA sequence and the BLAST 2.0 homology search tool.
Northern Blot Analysis-The following multiple tissue mRNA blots (MTNs) were used for Northern blot analysis: Human 12-lane MTN Blot, Human 12-lane MTN Blot II, and Human Endocrine System MTN Blot from CLONTECH, which contained a minimum of 1 g of polyadenylated RNA per lane. The blots were hybridized with ϳ1.0-kilobase pair 32 P-labeled cDNA probe in ExpressHyb hybridization solution (CLONTECH) according to the manufacturer's instructions. Briefly, the blots were prehybridized in ExpressHyb solution for 30 min at 68°C and transferred to a fresh solution containing 2 ϫ 10 6 cpm/ml denatured radiolabeled cDNA. The hybridization was performed at 68°C for 1 h. The blots were rinsed in 2ϫ SSC, 0.05% SDS several times at room temperature and washed in 0.1ϫ SSC, 0.1% SDS for 10 min at 50°C. The mRNA bands were visualized by exposure to x-ray film at Ϫ70°C with two intensifying screens for 1-10 days.
Expression in Sf9 Cells-Expression of the cDNA in Sf9 cells was performed essentially as described previously (20). To produce recombinant protein, Sf9 cells were infected at a virus/cell ratio of 10:1. Cells were collected after 3 days of incubation at 27°C, resuspended in 0.01 M potassium phosphate, pH 7.4, 0.25 M sucrose, 0.1 mM EDTA, 1 mM dithiothreitol, and protease inhibitors (1.5 g/ml aprotinin, 1.5 g/ml leupeptins, and 1.0 g/ml pepstatin A), and homogenized using a French pressure cell. The unbroken cells, cellular debris, and nuclei were removed by centrifugation at 1000 ϫ g for 10 min, and then mitochondria were removed by centrifugation at 10,000 ϫ g for 30 min. Microsomes were pelleted by centrifugation at 105,000 ϫ g for 1 h through a 0.6 M sucrose cushion and resuspended in 90 mM potassium phosphate, pH 7.4, 40 mM KCl, 0.1 mM EDTA, 1 mM dithiothreitol, and 20% glycerol. Protein concentration was determined by the method of Lowry et al. (30) using bovine serum albumin as a standard. Alkaline Extraction and Western Blot Analysis-Five-l aliquots of microsomes (1.8 g/l) containing the recombinant protein were incubated with 100 l of 100 mM sodium carbonate and 25 mM potassium acetate (extraction buffer) or with phosphate-buffered saline (PBS), pH 7.4, or with 1-10% Triton X-100 in PBS for 30 min on ice. After incubation, samples were loaded onto 100-l cushions of 0.5 M sucrose prepared in extraction buffer, PBS, or Triton/PBS and centrifuged for 1 h at 200,000 ϫ g. Pellets were dissolved in 20 l of SDS-polyacrylamide gel electrophoresis sample buffer. Supernatants were precipitated with an equal volume of ice-cold 50% trichloroacetic acid for 30 min on ice and centrifuged for 3 min at 12,000 ϫ g. The resulting pellets were washed twice with ethyl ether, dried, and dissolved in 20 l of SDS-polyacrylamide gel electrophoresis sample buffer. After separation in a 15% denaturing polyacrylamide gel, samples were transferred to Hybond-P membrane (Amersham Pharmacia Biotech). Protein was detected using the ECL Western blotting analysis system (Amersham Pharmacia Biotech) according to the manufacturer's instructions. Rabbit antibodies raised against the N-terminal fragment of RoDH-like 3␣-HSD were used as primary antibodies at a 1:3000 dilution in 3% bovine serum albumin, 20 mM Tris, pH 7.6, 137 mM NaCl, and 0.1% Tween 20. Visualization was performed using horseradish peroxidaseconjugated anti-rabbit antibodies (at a 1:10,000 dilution) and ECL Western blotting detection reagents (Amersham Pharmacia Biotech).
Identification of Reaction Products and Determination of Kinetic Constants-All reactions were performed in 90 mM potassium phosphate, pH 7.4, and 40 mM KCl at 37°C (reaction buffer) in siliconized glass tubes as described previously (20). Commercially available radiolabeled steroids (PerkinElmer Life Sciences; ϳ40 -60 Ci/mmol each) were diluted with cold steroids (Steraloids Inc. (Newport, RI) and Sigma) dissolved in dimethyl sulfoxide (Me 2 SO). The 500-l reactions (final concentration of Me 2 SO Ͻ 1%) were started with the addition of enzyme. After 15-60 min at 37°C, steroids were extracted and separated by development in toluene:acetone (4:1) on silica gel TLC plates (Sigma). TLC plates containing 3 H-labeled steroids were exposed to a PhosphorImager (Molecular Dynamics) tritium screen overnight and/or cut into 1-cm-wide sections, which were then counted in scintillation liquid (Bio-Safe II). Products of each reaction were identified by comparison with reference steroids. For determination of apparent K m values, six concentrations between 0.1 and 17.5 M were used for allopregnanolone and dihydrotestosterone; and concentrations between 0.5 and 25 M were used for androstanediol, androsterone, and dehydroepiandrosterone. The amount of product formed was less than 10% of the amount of substrate within the 15-min reaction time and was linearly proportional to the amount of microsomes added. A control without added cofactor was included with each experiment and subtracted from each experimental data point. The apparent K m values for oxidation of hydroxysteroids were determined at a fixed NAD ϩ concentration (1 mM), and the apparent K m value for reduction of dihydrotestosterone was determined at a fixed NADH concentration (1 mM). Each K m determination was repeated at least three times. The apparent K m values for cofactors were determined at a fixed saturating concentration of substrate with six concentrations of each cofactor (1-400 M for NAD ϩ and 1-200 M for NADH).
The activity with all-trans-retinol and 13-cis-retinol was determined in the same reaction buffer used for steroid assays. The reaction was allowed to proceed for 30 min at 37°C, and the reaction products were extracted and analyzed by HPLC as described previously (21).
Coupled in Vitro Transcription/Translation and Protease Protection Assay-The coding region of the cDNA was cloned into BamHI/EcoRI restriction sites of expression vector pT7/T3-19 (Ambion Inc., Austin, FIG. 2. cDNA sequence and deduced protein sequence of human microsomal 3␣-HSD. Numbers on the right correspond to nucleotide sequence (italic) and amino acid sequence. The termination codon is indicated by an asterisk. The sequences of primers used to obtain the 5Ј-RACE PCR product are underlined by arrows.
TX) under the control of T7 promoter. The cDNA for 11␤-HSD1 (31) was amplified from 5Ј-RACE-ready human liver cDNA using Pfu polymerase and primers GCTGGATCCGCCATGGCTTTTATGAAAAA-ATATCTC (sense primer) and AGGTCTAGACTACTTGTTTATGAAT-CTGTCC (antisense primer), which contained BamHI and XbaI restriction sites (underlined), respectively. The PCR product of expected size was cloned into pT7/T3-18 vector cleaved with BamHI and XbaI restriction endonucleases and sequenced to verify the fidelity of PCR amplification. The expression constructs were subjected to in vitro transcription by T7 RNA polymerase and translation in reticulocyte lysate in the presence or absence of dog pancreas microsomes (TNT Quick system; Promega) according to the manufacturer's instructions. In a typical assay, 0.5-1 g of plasmid DNA, 20 Ci of [ 35 S]methionine (Amersham Pharmacia Biotech), and 0.5-1 l of canine pancreatic microsomal membranes (Promega) were incubated for 60 -90 min at 30°C in a final volume of 12.5 l. Translocation of polypeptides to the lumenal side of the microsomes was assayed by proteinase K treatment. Equal aliquots of the in vitro transcription/translation product were incubated for 60 min on ice with or without 200 g/ml proteinase K (Roche Molecular Biochemicals) in a final volume of 5 l. Proteinase K was inactivated by the addition of 4 mM phenylmethylsulfonyl fluoride. Proteins were subjected to 12% SDS-polyacrylamide gel electrophoresis and analyzed by autoradiography.
Thirty g of 3␣-HSD-containing microsomes from Sf9 cells were incubated with or without proteinase K in the presence or absence of 1% CHAPS. Reactions were incubated at 37°C for 15 min with various amounts of protease relative to microsomal protein (g:g) in a final reaction volume of 36 l. The reaction was stopped by the addition of protease inhibitor phenylmethylsulfonyl fluoride to the final concentration of 2 mM. Samples of the reaction mixture were analyzed by Western blotting and by activity measurements using 5 M androsterone as substrate.

Characterization of the Primary Structure and Genomic
Organization-The 1513-base pair cDNA shown in Fig. 2 was constructed from the overlapping sequences of the original clone W17165 (nucleotides 346 -1513) and the 5Ј-RACE PCR product (nucleotides 1-449) obtained in this study (see "Experimental Procedures"). The open reading frame starts at nucleotide 154 and ends in a stop codon at nucleotide 1,113, resulting in a 319-amino acid-long polypeptide (Fig. 2). A search of the GenBank TM data base for homologous sequences revealed that three unpublished sequences exhibit similarity to the cDNA reported here: (a) human retinol dehydrogenase homolog gene (AF067174), (b) Homo sapiens retinol dehydrogenase homolog isoform-1 mRNA (AF240698), and (c) H. sapiens retinol dehydrogenase homolog isoform-2 mRNA (AF240697). The first cDNA is identical to segments 147-603 and 667-1513 in the cDNA shown in Fig. 2 but is missing a segment coding for 21 amino acids between Pro-150 and Val-172. Instead, in this position, it contains nucleotides 1-45 of the 5Ј-untranslated sequence joined with an unidentified 31-base pair fragment. Retinol dehydrogenase homolog isoform-1 lacks 120 base pairs coding for 40 amino acids between Glu-204 and Ser-246, whereas isoform-2 has an insertion of 113 nucleotides between positions 94 and 95 in the 5Ј-untranslated region (Fig. 2) as well as four mismatched nucleotides and one missing nucleotide, which causes a frameshift.
To obtain the full-length cDNA and further confirm the cDNA and deduced protein sequence, we performed PCR amplification of mRNA isolated from two human tissues, liver and heart, using gene-specific primers flanking the coding region. Sequencing of the PCR products showed that they were identical with the sequence in Fig. 2. Furthermore, a BLAST 2.0 search of the Human Genome data base revealed that the cDNA sequence determined in this study (AF343729) is 100% identical to five consecutive segments in the NCBI-assembled contig NT 005343 assigned to chromosome 2q31.1 (Fig. 3A). The corresponding gene spans over 26 kilobase pairs and does not appear to have pseudogenes. The proposed exon-intron boundaries follow the canonical GT/AG rule and are summarized in Table I.
The deduced protein sequence of the novel protein contains the signature cofactor binding motif G(X) 3 GXG at Gly-36 and the active site consensus sequence Y(X) 3 K at Tyr-176 characteristic of the short chain dehydrogenases/reductases. The amino acid sequence is most closely related to retinol/sterol dehydrogenases of the short chain dehydrogenase/reductase superfamily: (a) RoDH-4, 47% identity; (b) RoDH-like 3␣-HSD, 43% identity; and (c) Rdh5, 44% identity. Furthermore, the positions of exon-intron junctions in the translated region of the novel gene and the sizes of exons 3 and 4 are identical to those in the genes for RoDH-4, RoDH-like 3␣-HSD, and Rdh5 (Fig. 3B). However, the new gene is present on chromosome 2q31.1, whereas retinol/sterol dehydrogenases are clustered on chromosome 12q13 (28).
Tissue Distribution-Tissue distribution of the corresponding message was examined by Northern blot analysis. Hybridization of pre-made Northern blots with radiolabeled cDNA revealed that the message for the putative short chain dehydrogenase is present in a number of human tissues (Fig. 4). The most intense signal was observed in trachea. Relatively strong signals were detected in spinal cord, bone marrow, heart, and colon. Weaker bands were also present in lymph node, brain, lung, placenta, testis, prostate, and mammary gland. A total of three different-size mRNA species were detected, which exhibited tissue-specific expression. The longest mRNA species were detected in trachea, testis, and lung. An intermediate size mRNA was present in trachea, colon, placenta, lung, bone marrow, and lymph node, with trace amounts seen in mammary gland, prostate, and stomach. The shortest mRNA was detected in spinal cord, lymph node, brain, bone marrow, and heart.
Expression and Characterization of the Recombinant Enzyme-To establish whether the novel cDNA codes for a functional enzyme, we expressed the corresponding protein using the baculovirus expression system. Similar to human retinol/ sterol dehydrogenases, the recombinant protein was present in the microsomal fraction of Sf9 cells as indicated by Western blot analysis using antibodies against the conserved N-terminal cofactor-binding domain of retinol/sterol dehydrogenases. Under the same conditions, no protein bands were detected in microsomes isolated from Sf9 cells infected with wild-type virus, which did not express the recombinant protein (data not shown). To determine whether the new short chain dehydrogenase is an integral membrane protein, microsomal fraction from Sf9 cells that expressed the recombinant protein was subjected to alkaline and detergent extractions. Equal amounts of microsomes were incubated with either sodium carbonate buffer, pH 11.5 (alkaline extraction), Triton X-100/PBS (detergent extraction), or PBS, pH 7.4 (control) (Fig. 5). Extracted proteins were separated from membrane-bound proteins by ultracentrifugation and analyzed by Western blotting. As shown in Fig. 5, the recombinant protein remained membranebound after alkaline extraction but was partially solubilized by increasing concentration of a detergent, consistent with the behavior of an integral membrane protein.
Next, we tested whether the putative short chain dehydrogenase is enzymatically active. Because its amino acid sequence showed the closest similarity to retinol/sterol dehydrogenases, we examined whether it possessed a hydroxysteroid dehydrogenase activity using reaction conditions established previously for RoDH-4 (20) and RoDH-like 3␣-HSD (21). Microsomes were incubated with a number of tritiated steroid compounds for 1 h at 37°C in the presence of 1 mM NAD ϩ or NADH. The reaction products were extracted, separated by TLC, and visualized using a PhosphorImager. At a 5 M concentration of each substrate, the highest percentage of conversion in the oxidative direction in the presence of NAD ϩ was observed with 3␣-hydroxysteroids, 3␣-androstanediol and allopregnanolone (Fig. 6A). 3␣-Androstanediol contains two hydroxyl groups in positions 3␣ and 17␤ (Fig. 1). The enzyme possessed both 3␣-HSD and 17␤-HSD activity, as evidenced by the appearance of androstanedione with ketone groups in positions 3 and 17 (Fig. 6A). To estimate the relative efficiency of the two activities, we used substrates that have only one hydroxyl group: 3␣-hydroxyl (androsterone) or 17␤-hydroxyl (di-hydrotestosterone) (Fig. 1). The 3␣-hydroxyl group was oxidized much more efficiently (androsterone to androstanedione) than the 17␤-hydroxyl group (dihydrotestosterone to androstanedione). To evaluate the 3␤-HSD activity of the enzyme, we used dehydroepiandrosterone as substrate. As seen in Fig. 6A, no oxidation products of dehydroepiandrosterone were detected by autoradiography, indicating that the rate of conversion was not sufficient to detect the respective oxidation product.
One of the best substrates for the novel 3␣-HSD in the oxidative direction was allopregnanolone (3␣-tetrahydroprogesterone) (Fig. 6A). Interestingly, besides a significant amount of dihydroprogesterone with ketone group in position 3, an additional product, isopregnanolone (3␤-tetrahydroprogesterone), appeared in this reaction. This implied that the ketone group on carbon 3 of dihydroprogesterone could be reduced to either a 3␣ or 3␤-hydroxyl group. This type of activity was observed previously during prolonged incubations of RoDH-like 3␣-HSD with allopregnanolone and NAD ϩ (22).
Analysis of the reaction products in the reductive direction in the presence of NADH showed that ketone group on carbon 3 was reduced to hydroxyl group with the highest catalytic effi-

New Type of Human Microsomal 3␣-HSD
ciency (dihydrotestosterone to 3␣/3␤-androstanediol) (Fig. 6B). The 17-ketone group on androsterone was also reduced, albeit at a much slower rate. To ensure that the observed 3␤-and 17␤-hydroxysteroid dehydrogenase activity was not due to endogenous activity of insect cell microsomes, we determined the activity of microsomes isolated from Sf9 cells that were infected with wild-type virus and did not express 3␣-HSD. At a 5 M substrate concentration, 0.2 pmol of androstanedione and 0.08 pmol of androstendione were formed by 3 g of control microsomes from dihydrotestosterone and dehydroepiandrosterone, respectively, in the presence of NAD ϩ . In contrast, the same amount of microsomes that contained 3␣-HSD produced 199 pmol of androstanedione and 28 pmol of androstendione. The endogenous hydroxysteroid activity of Sf9 microsomes from the cells infected with wild-type virus was also tested using [ 14 C]dihydrotestosterone in the presence of NAD ϩ (17␤-hydroxysteroid dehydrogenase activity) or NADH (3-keto reductase activity). As shown in Fig. 6C, no products were detected in the oxidative or the reductive direction. Based on this analysis, we concluded that all of the observed steroid conversions were catalyzed by the novel 3␣-HSD. Before measuring kinetic constants for oxidation and reduction of steroids, we determined whether NAD ϩ and NADH were, in fact, the preferred cofactors. In the oxidative direction, using 5 M allopregnanolone as substrate, the reaction rate was 44-fold higher in the presence of 1 mM NAD ϩ than in the presence of 1 mM NADP ϩ . In the reductive direction, using 10 M dihydrotestosterone as substrate, the rate was 9-fold higher with 1 mM NADH than it was with 1 mM NADPH. Thus, NAD ϩ and NADH were the preferred cofactors. The apparent K m value for NAD ϩ determined with 20 M allopregnanolone was 72.0 Ϯ 5.0 M. The apparent K m value for NADH determined with 25 M dihydrotestosterone was 9.0 Ϯ 0.5 M.
Considering that multiple products are formed by the enzyme over long periods of incubation (or with high enzyme concentrations), we established conditions under which only one reaction product was formed. These conditions were used to determine kinetic constants for steroid substrates. Specifically, the apparent K m and V max values for allopregnanolone and 3␣-androstanediol were measured when there was no detectable formation of secondary products (isopregnanolone and androstanedione, respectively). Consistent with autoradiography analysis of the reaction products, allopregnanolone and androstanediol were the best substrates in the oxidative direction with the apparent K m values of 5-7 M (Table II). The apparent K m value for androsterone was ϳ5-fold higher (Table II). Kinetic analysis also showed that the enzyme was capable of binding 3␤-hydroxysteroid dehydroepiandrosterone; however, the rate of conversion was about 20-fold lower compared with that of allopregnanolone (Table II), which explains why the corresponding product could not be visualized using a Phos-phorImager (Fig. 6A). Dihydrotestosterone (3-ketone group) FIG. 4. Northern blot analysis of 3␣-HSD distribution in human tissues. Human multiple tissue Northern blots (CLONTECH) containing a minimum of 1 g polyadenylated RNA/lane were hybridized with human radiolabeled 3␣-HSD cDNA at high stringency conditions. The blots were exposed to x-ray film for 1 day (the panel with uterus and trachea on the far left) and then re-exposed for 10 days (all other panels). The hybridized blots were stripped of residual radioactivity, reprobed with ␤-actin cDNA, and exposed for 3 h (bottom panel). Positions of the size standards are indicated on the left.

FIG. 5. Alkaline extraction and Western blot analysis of recombinant 3␣-HSD.
Microsomal membranes containing 3␣-HSD were extracted with 10% Triton X-100 in PBS (10% Tx-100), 1% Triton X-100 in PBS (1% Tx-100), sodium carbonate buffer, pH 11.5 (alkaline extraction), or PBS, pH 7.4 (PBS). Solubilized proteins (S) were separated from integral membrane proteins (P) by ultracentrifugation. Distribution of 3␣-HSD between the soluble and the membrane-bound fractions was analyzed by Western blotting as described under "Experimental Procedures. "   FIG. 6. Analysis of the reaction products produced by 3␣-HSD in the presence of NAD ϩ (A) or NADH (B). Control reactions performed in the absence of cofactor (ϪNAD ϩ /ϪNADH) are shown next to the sample with the same substrate in the presence of cofactor. All reactions were allowed to proceed for 1 h at 37°C. Substrates were used at a concentration of 5 M. The microsomes were used at concentration of 25 g/ml in the reaction volume. C, analysis of the endogenous activity of Sf9 microsomes isolated from cells infected with wild-type baculovirus toward dihydrotestosterone in the presence of NAD ϩ (ϩ NAD ϩ ) (17␤dehydrogenase activity) or NADH (ϩ NADH) (3 keto-reductase activity). ADT, androsterone; DHT, dihydrotestosterone; ADIOL, 3␣-androstanediol; DHEA, dehydroepiandrosterone; ALLO-P, allopregnanolone; ISO-P, isopregnanolone; DHP, dihydroprogesterone; DIONE, androstanedione.
was reduced to androstanediol with a catalytic efficiency similar to that for the oxidation of 3␣-hydroxyl group (Table II).
To determine whether the new enzyme was active toward retinoids, we incubated 30 -300 g of microsomal membranes containing the recombinant protein with 10 -50 M all-transretinol or 13-cis-retinol in the presence of 1 mM NAD ϩ . After a 30-min incubation at 37°C, the reaction products were extracted and analyzed by HPLC as described previously (21). The amount of retinaldehyde formed was calculated from the calibration curve of the amount of pure retinal injected onto the column. Based on these measurements, 71 pmol were produced by 30 g of microsomal protein over the 30-min incubation time from 10 M retinol. For comparison, 9600 pmol of dihydroprogesterone would have been produced from 10 M allopregnanolone by the same amount of protein under the identical conditions. Considering that the rate of retinol oxidation is at least 100 times lower than that of allopregnanolone, the new enzyme is not efficient as retinol dehydrogenase.
Transmembrane Topology of 3␣-HSD-Because we have established that the novel 3␣-HSD is an integral membrane protein, we were interested in determining its transmembrane orientation. Analysis of its protein sequence for signature sites and motifs indicated that there is a potential N-terminal transmembrane segment (amino acids 2-17) as well as three potential N-glycosylation sites at Asn-161, Asn-187, and Asn-253. Because N-glycosylation is carried out exclusively on the luminal side of the endoplasmic reticulum, the appearance of protein species with higher subunit molecular weight in the presence of microsomes would indicate that they become glycosylated and were therefore exposed to the lumen.
To investigate the transmembrane topology of 3␣-HSD, we used a coupled in vitro transcription/translation system and canine pancreatic microsomal membranes. 11␤-HSD1, a protein with known lumenal orientation and three glycosylation sites (32), served as a positive control for glycosylation. 3␣-HSD was synthesized in vitro in the presence or absence of canine microsomes. Analysis of the reaction products showed that the size of 3␣-HSD produced in the absence of microsomes was identical to that produced in the presence of microsomes, indicating that the three glycosylation sites in 3␣-HSD were not glycosylated (Fig. 7A). At the same time, all three sites were glycosylated in the positive control, 11␤-HSD1 (Fig. 7A). Then, the reaction products were treated with proteinase K to determine whether the protein was translocated across the membrane and protected against the protease (see "Experimental Procedures"). In agreement with the lack of glycosylation, 3␣-HSD produced in the presence of microsomes was not protected from proteinase K, whereas the glycosylated 11␤-HSD1 was fully protected. This outcome is consistent with the cytosolic orientation of 3␣-HSD, in contrast to the lumenal orientation established for 11␤-HSD1 (32).
To obtain additional evidence for the cytosolic orientation of the new 3␣-HSD, we incubated enzyme-containing microsomes from Sf9 cells with different concentrations of proteinase K in the presence or absence of 1% CHAPS and analyzed the residual activity and the amount of protein left after protease treatment. 3␣-HSD was readily proteolysed by proteinase K in both the presence and absence of CHAPS, suggesting that the disruption of membrane integrity was not essential for proteolysis to occur (Fig. 7B). The activity of the enzyme incubated with proteinase K decreased from 1.87 nmol/min/mg microsomal protein (100%) to 0.86 nmol/min/mg microsomal protein (46%) at a 1:250 protease:microsomes ratio (g:g) (Fig. 7C). No change in either 3␣-HSD activity or protein was observed in the absence of proteinase K. Based on the above experimental data, we conclude that the new short chain dehydrogenase is an integral membrane protein, which faces the cytosolic side of the microsomal membrane.

DISCUSSION
The novel 3␣-HSD characterized in the present study exhibits less than 50% amino acid sequence identity to any known protein. It is most closely related to retinol/sterol dehydrogenases of the short chain dehydrogenase/reductase superfamily.  Protease protection assay of 3␣-HSD produced in vitro using coupled transcription/translation system (A) and of recombinant 3␣-HSD in Sf9 cell-derived microsomes (B). A, proteins (3␣-HSD and 11␤-HSD1) were produced in the presence (ϩ) or absence (Ϫ) of canine pancreatic microsomes (Ms) and incubated with (ϩ) or without (Ϫ) 200 g/ml proteinase K (PK). Proteins were separated by 12% SDS-polyacrylamide gel electrophoresis, and dried gels were subjected to autoradiography. B, microsomal membranes isolated from Sf9 cells that express 3␣-HSD (Prot) were incubated with proteinase K (PK) at various protein:protein ratios in the presence (ϩ) or absence (Ϫ) of 1% CHAPS (CHAPS). Samples of the reactions were analyzed by Western blotting and activity assays (C). C, the residual 3␣-HSD activity was determined using 5 M androsterone as substrate. 100% activity corresponds to the activity of 3␣-HSD incubated under the same conditions in the absence of proteinase K. Furthermore, the gene for 3␣-HSD shares similar structural organization with the genes for retinol/sterol dehydrogenases (28). Some variation is observed in the length of exon 5, which is 1 amino acid longer that the corresponding exon in the Rdh5 gene and 2 amino acids longer than that in the RoDH-4 and RoDH-like 3␣-HSD genes. Interestingly, 3␣-HSD gene is found on a different chromosome and does not appear to have satellite pseudogenes like RoDH-4 and RoDH-like 3␣-HSD genes (28), which might indicate that it appeared later in evolution.
Analysis of the sequences deposited in the GenBank TM expressed sequence tag data base showed that cDNA fragments identical to different segments of 3␣-HSD cDNA were isolated from multiple sources: lymph node (nine reports), testis (three reports), normal colon and colon adenocarcinoma (three reports), cervical tumor (one report), breast (one report), uterus and endometrial adenocarcinoma (four reports), fetal lung (one report), pancreas adenocarcinoma (two reports), and lymphoma (one report). Thus, 3␣-HSD appears to exhibit wide tissue distribution. Indeed, Northern blot analysis of human tissues revealed that 3␣-HSD mRNA is present in the central nervous system and in a number of peripheral endocrine tissues. Interestingly, the size of the predominant mRNA species varied depending on the tissue source, suggesting that there might be a tissue-specific splicing of 3␣-HSD mRNA. In fact, we found that the 113-nucleotide segment inserted in the 5Ј-untranslated sequence of a nearly identical clone submitted under the name of retinol dehydrogenase homolog isoform-2 (AF240697) is located within the 11,805-base pair sequence of putative intron 1. This observation suggests that alternatively spliced products may exist.
The most striking difference in the distribution pattern of the novel 3␣-HSD compared with that of retinol/sterol dehydrogenases is that the 3␣-HSD message is practically absent in liver. The lack of hybridization with the liver mRNA serves as a confirmation that the cDNA probe does not cross-hybridize with either RoDH-4 or RoDH-like 3␣-HSD mRNA, both of which are very abundant in the liver.
The presence of 3␣-HSD in the brain is consistent with its activity toward allopregnanolone, which serves as a potent allosteric regulator of ␥-aminobutyric acid type A receptors. Allopregnanolone is likely to be produced in the brain by cytosolic 3␣-HSDs, AKR1C1 and AKR1C2, which catalyze the reduction of the 3-ketone group on dihydroprogesterone (13). The back-oxidation of allopregnanolone to dihydroprogesterone is thought to be catalyzed by an unidentified membrane-bound NAD ϩ -dependent 3␣-HSD (4,5). Previously, we have reported that RoDH-like 3␣-HSD is capable of oxidizing allopregnanolone and is expressed in various regions of the human brain and in the spinal cord (21), suggesting that it may function as allopregnanolone dehydrogenase in the central nervous system. This study shows that, in addition to RoDH-like 3␣-HSD, the central nervous system contains a second isoform of microsomal oxidative 3␣-HSD active toward allopregnanolone. Interestingly, similar to RoDH-like 3␣-HSD (21,22), the newly characterized enzyme exhibits "epimerase" activity toward allopregnanolone, slowly converting it to 3␤-hydroxysteroid isopregnanolone in the presence of NAD ϩ . We have previously proposed that NADH, which is produced during oxidation of allopregnanolone, might be retained in the active site and trigger the nonstereospecific reduction of the 3-ketone group on dihydroprogesterone to both a 3␣-hydroxyl and a 3␤-hydroxyl group. Consistent with this hypothesis, the new 3␣-HSD exhibits a higher affinity for NADH than for NAD ϩ (the apparent K m value is 9 versus 72 M), similar to RoDH-like 3␣-HSD (0.18 versus 1.9 M) (21). At the same time, RoDH-4, which has almost identical K m values for NAD ϩ (0.13 M) and NADH (0.1 M), does not seem to possess the epimerase activity. 2 It should also be noted that, in addition to allopregnanolone, RoDH-like 3␣-HSD epimerized androsterone into epiandrosterone. However, we did not observe epimerization of androsterone with the new 3␣-HSD, possibly due to the 5-fold higher apparent K m value of the enzyme for androsterone compared with allopregnanolone.
Besides utilizing allopregnanolone as substrate, the new 3␣-HSD is equally efficient at converting 3␣-androstanediol into a potent androgen, dihydrotestosterone. This is consistent with the expression of the enzyme in testis and prostate. Previously, Biswas and Russell (15) proposed that RoDH-like 3␣-HSD, which is expressed in prostate and testis, contributes to the back-oxidation of 3␣-androstanediol to dihydrotestosterone in these tissues. This study shows that, in addition to RoDH-like 3␣-HSD, dihydrotestosterone can be produced in testis and prostate by the novel isoform of 3␣-HSD.
Similar to retinol/sterol dehydrogenases, 3␣-HSD is an integral membrane protein. Therefore, depending on its transmembrane topology, the active site of the enzyme may be exposed either to the cytosol or to the lumen of the endoplasmic reticulum. This, in turn, may affect its access to the substrates and cofactors. Previous analysis of the transmembrane topology of Rdh5, which was proposed to catalyze the biosynthesis of 11cis-retinaldehyde in retinal pigment epithelium, showed that Rdh5 is anchored to the membranes of smooth endoplasmic reticulum by two hydrophobic peptide segments (23). The catalytic domain of this enzyme is confined to the lumenal compartment, suggesting that generation of 11-cis-retinaldehyde occurs in the lumen of the endoplasmic reticulum (23). In contrast to Rdh5, 3␣-HSD appears to be facing the cytosolic side of the membrane, indicating that it will utilize the cytosolic pool of steroids and nucleotides. In the liver cytosol, and presumably in other tissues, the NAD ϩ :NADH ratio is about 1000, whereas the NADP ϩ :NADPH ratio is about 0.01 (33). This suggests that the NAD ϩ -preferring oxidoreductases, which face the cytosol like the new 3␣-HSD, will function in the oxidative direction. The available substrate and cofactor pool for the lumenally oriented Rdh5, which also prefers NAD ϩ over NADP ϩ as cofactor (34), is less clear.
The existence of enzymes that share similar substrates but exhibit different transmembrane orientation is not unusual. For example, two other members of the short chain dehydrogenase/reductase superfamily, 11␤-HSD type 1 and type 2, which catalyze the interconversion between a potent glucocorticoid cortisol and a weak glucocorticoid cortisone, exhibit opposite transmembrane orientation: 11␤-HSD type 1 is a lumenally oriented glycoprotein, whereas 11␤-HSD type 2 faces the cytosol (32). Interestingly, these two enzymes exhibit a different cofactor preference: 11␤-HSD type 1 catalyzes oxidation and reduction using NADP(H) as cofactor, whereas 11␤-HSD type 2 is NAD ϩ -specific and catalyzes only 11␤-dehydrogenation. This is in contrast to Rdh5 and 3␣-HSD, which both appear to exhibit a cofactor preference for NAD ϩ .
Besides the difference in the transmembrane topology and tissue distribution, 3␣-HSD exhibits different substrate specificity compared with Rdh5 and the all-trans-retinol-oxidizing microsomal dehydrogenases, RoDH-4 and RoDH-like 3␣-HSD. Rdh5 has similar affinity for retinoids (K 0.5 of 6.3-6.6 M for 9-cis and 11-cis-retinol) and 3␣-androstanediol (K 0.5 of 6.4 M) (16) and is about two times more efficient as retinol dehydrogenase than as a steroid dehydrogenase. RoDH-4 and RoDHlike 3␣-HSD occupy an intermediate position and are more efficient as 3␣-hydroxysteroid dehydrogenases with apparent K m values of ϳ0.2 M for 3␣-hydroxysteroids (20,21) while retaining the retinol-oxidizing capacity (K m value of 3.2 M) (21). The new 3␣-HSD is practically inactive with retinoid compounds and is most efficient as allopregnanolone and 3␣androstanediol dehydrogenase, representing the opposite end of the spectrum. Thus, experimental data obtained in this study suggest that the new member of the short chain dehydrogenase/reductase superfamily is a novel type of microsomal NAD ϩ -dependent 3␣-HSD with unique tissue distribution, transmembrane topology, and catalytic properties.