Aldo-keto Reductase 1B15 (AKR1B15)

Background: Aldo-keto reductases (AKRs) are enzymes involved in the metabolism of carbonyl substrates. Results: Two alternatively spliced protein isoforms encoded by the human AKR gene AKR1B15 were identified. The AKR1B15.1 isoform catalyzes reduction of steroids and 3-keto-acyl-CoA conjugates and localizes to mitochondria. Conclusion: AKR1B15.1 is a mitochondrial carbonyl reductase. Significance: AKR1B15.1 is a new enzyme with unique localization and catalytic features. Aldo-keto reductases (AKRs) comprise a superfamily of proteins involved in the reduction and oxidation of biogenic and xenobiotic carbonyls. In humans, at least 15 AKR superfamily members have been identified so far. One of these is a newly identified gene locus, AKR1B15, which clusters on chromosome 7 with the other human AKR1B subfamily members (i.e. AKR1B1 and AKR1B10). We show that alternative splicing of the AKR1B15 gene transcript gives rise to two protein isoforms with different N termini: AKR1B15.1 is a 316-amino acid protein with 91% amino acid identity to AKR1B10; AKR1B15.2 has a prolonged N terminus and consists of 344 amino acid residues. The two gene products differ in their expression level, subcellular localization, and activity. In contrast with other AKR enzymes, which are mostly cytosolic, AKR1B15.1 co-localizes with the mitochondria. Kinetic studies show that AKR1B15.1 is predominantly a reductive enzyme that catalyzes the reduction of androgens and estrogens with high positional selectivity (17β-hydroxysteroid dehydrogenase activity) as well as 3-keto-acyl-CoA conjugates and exhibits strong cofactor selectivity toward NADP(H). In accordance with its substrate spectrum, the enzyme is expressed at the highest levels in steroid-sensitive tissues, namely placenta, testis, and adipose tissue. Placental and adipose expression could be reproduced in the BeWo and SGBS cell lines, respectively. In contrast, AKR1B15.2 localizes to the cytosol and displays no enzymatic activity with the substrates tested. Collectively, these results demonstrate the existence of a novel catalytically active AKR, which is associated with mitochondria and expressed mainly in steroid-sensitive tissues.

The aldo-keto reductase (AKR) 3 superfamily comprises 15 families containing over 150 members that are present in all phyla (1,2). AKRs are multifunctional enzymes that catalyze the reduction of biogenic and xenobiotic aldehydes and ketones as well as the synthesis and metabolism of sex hormones. The majority of AKRs catalyze oxidation-reduction reactions between carbonyl and alcohol groups, whereas enzymes of the AKR1D family reduce double bonds in the bile acid biosynthesis pathway, acting as 5␤-reductases (3). Some AKR proteins have very low or no activity and perform predominantly non-catalytic functions (e.g. structural (lens -crystallines: AKR1C10a and AKR1C10b) or regulatory and chaperone-like (voltage-gated potassium channel ␤-subunits of the AKR6 family: Kv␤) functions) (4,5).
The AKR1 family is the most numerous and has been further divided into five subfamilies (A-E). The AKR1B subfamily has been intensely studied due to the potential role of its founding member human aldose reductase (AKR1B1) in the development of diabetic complications (17,18). Under hyperglycemic conditions, AKR1B1 converts excess glucose into sorbitol, leading to osmotic and redox imbalances and resulting in tissue injury associated with diabetes (19 -21). Inhibition of AKR1B1 has been shown to prevent, delay, or reverse tissue injury due to hyperglycemia (19,22). In this context, a large number of studies and clinical trials have been devoted to finding efficient inhibitors of aldose reductase to prevent the development of diabetic complications; however, these efforts have met with limited success due to problems in trials design, low efficacy, and nonspecific side effects of inhibitors (17,23). In addition to glucose, AKR1B1 catalyzes the reduction of several substrates of physiological significance, including advanced glycation end-product precursors, 4-hydroxy-trans-2-nonenal, and oxidized phospholipids (11,24,25), and has been suggested to play important roles in the development of atherosclerosis (26), ischemic preconditioning (27), and restenosis (28).
AKR1B1 is closely related to the small intestine aldose reductase (AKR1B10) (29,30). In contrast to AKR1B1, AKR1B10 is expressed mainly in small intestine, colon, liver, thymus (29), and adrenal gland (30). AKR1B10 shares 71% amino acid sequence identity with AKR1B1 and exhibits substrate specificity similar to aldose reductase with the exception that it has significantly higher catalytic efficiency with all-trans-retinal (31). AKR1B10 is strongly overexpressed in lung and hepatic carcinomas (squamous cell and adeno-carcinomas) (29) as well as in colorectal and uterine cancers (32) and has been implicated in conferring resistance to anticancer drugs (33,34).
Recently, a novel gene, AKR1B15, with 91% identity to AKR1B10 has been predicted in the genetic cluster encompassing AKR1B1 and AKR1B10 on human chromosome 7. We previously reported that this gene encodes a functional protein (35). However, in contrast to AKR1B1 and AKR1B10, the enzymatic activity of this newly identified AKR was low, and the protein expressed with an N-terminal His tag was found in the microsomal fraction in both the mammalian and bacterial expression systems (35). Although orthologs of AKR1B1 are known in rodents (AKR1B3 in mouse and AKR1B4 in rat), direct orthology between AKR1B15 and AKR1B10 and rodent AKR1Bs has not been established so far.
In the present study, we show that the AKR1B15 gene gives rise to two alternatively spliced mRNA products, each coding for a unique protein, hereafter referred to as AKR1B15.1 and AKR1B15.2. Furthermore, we characterize the catalytic activity, tissue distribution, and subcellular localization of both AKR1B15 isoforms.
Enrichment of Mitochondria from BeWo Cells-Mitochondria were enriched from the BeWo cell line using a pump-controlled cell rupture system, following the protocol published by Schmitt et al. (37). For cell rupture, 2 ϫ 10 7 freshly harvested BeWo cells were resuspended in 4 ml of isolation buffer (300 mM sucrose, 5 mM TES, and 200 M EGTA, pH 7.2) and passed three times through a cell homogenizer with 10-m clearance (Isobiotech) at a constant flow of 700 l/min. Ruptured cells were centrifuged at 800 ϫ g and 4°C for 5 min. The resulting supernatant was centrifuged once more at 9000 ϫ g and 4°C for 10 min. The concentration of total protein in the fractions (800 ϫ g pellet, 9000 ϫ g pellet, and 9000 ϫ g supernatant) was determined via the Bio-Rad DC protein assay kit (Bio-Rad).
Generation of Polyclonal and Monoclonal Antibodies against AKR1B15-The peptide sequence corresponding to a region with the most dissimilarity between AKR1B15 and AKR1B10 ( Fig. 1B, red box) was used as antigen for the generation of polyclonal IgG antibodies against AKR1B15 in rabbits, which was carried out by 21st Century Biochemicals (Marlboro, MA). The peptide Ac-NWRAFDFKEFSHLC-amide was synthesized, conjugated to an immune carrier, and used for the immunization of two rabbits to produce polyclonal antisera. Prior to use, the resulting antibodies were affinity-purified via the peptides deployed for immunization.
For the generation of the monoclonal antibody, the peptide comprising the amino acid sequence QGFKTGDDFFPKDDK-GNMISGKGTF from the human AKR1B15 protein (Fig. 1B, green box) was synthesized and coupled to ovalbumin (Peps4LS, Heidelberg, Germany). Lou/c rats were immunized subcutaneously and intraperitoneally with a mixture of 50 g peptideovalbumin, 5 nmol of CPG oligonucleotide (Tib Molbiol, Berlin), 500 l of PBS, and 500 l of incomplete Freund's adjuvant. A boost without adjuvant was given 6 weeks after the primary injection. Fusion of the myeloma cell line P3X63Ag8.653 (ATCC; CRL-1580 TM ) with rat immune spleen cells was performed using standard procedures. Hybridoma supernatants were tested in a differential ELISA with the biotinylated AKR1B15 peptide and an irrelevant biotinylated peptide on avidin-coated ELISA plates. mAbs that reacted specifically with the AKR1B15 peptide were further analyzed by Western blot.
Hybridoma culture supernatant of the anti-AKR1B15 clone 9A5 (rat IgG2a subclass) was used in this study.
SDS-PAGE and Western Blotting-Denatured proteins were loaded onto a 12% Mini-Protean or Criterion TGX gel (Bio-Rad) and separated by conventional electrophoresis in running buffer (25 mM Tris, pH 8.3, 192 mM glycine, 0.1% (m/v) SDS). Afterward, gels were either stained via Coomassie Brilliant Blue (0.05% (m/v) Coomassie Brilliant Blue R250, 10% (v/v) acetic acid, 40% (v/v) methanol) or blotted onto a PVDF membrane (Immobilon FL, Millipore) via semidry blot in transfer buffer (48 mM Tris, 39 mM glycine, 0.0375% (m/v) SDS, 20% (v/v) methanol). For testing antibody specificity, membranes after transfer were blocked with 5% skimmed milk powder in PBS and then incubated overnight with primary antibodies in 0.5% skimmed milk powder in PBS at 4°C. Membranes were washed three times with PBS for 10 min, followed by incubation with HRP-conjugated secondary antibodies (also in 0.5% skimmed milk powder in PBS) and another washing step. Signals were detected by incubating the membranes in Pierce ECL Plus Western blotting substrate (Thermo Scientific) according to the manufacturer's protocol and visualized using a Fusion FX7 system (Vilber Lourmat). A similar procedure was used to detect endogenous AKR1B15 isoforms, except that IR dye-labeled secondary antibodies and an Odyssey infrared imaging system (LI-COR) were used. Briefly, membranes were blocked with 50% Odyssey blocking buffer (for PBS) in PBS after transfer, antibodies were diluted in 50% Odyssey Blocking Buffer (for PBS) in PBS-T (0.05% Tween 20 in PBS), and washing steps were performed with PBS-T.
RNA Isolation and cDNA Synthesis-RNA from cultured cells was isolated using the RNeasy minikit (Qiagen) combined with a DNase I (Qiagen) digestion treatment. 1 g of RNA was reverse transcribed using oligo(dT) 18 primers and the Affini-tyScript qPCR cDNA synthesis kit (Agilent) or avian myeloblastosis virus reverse transcriptase (Promega) according to the manufacturer's protocols.
RT-PCR and qPCR-End point and real-time RT-PCR were carried out both with the same set of transcript-specific primer pairs (Table 1). For the end point RT-PCR, DreamTaq Green DNA polymerase (Thermo) was used according to the manufacturer's protocol, with 38 amplification cycles for AKR1B transcripts and 24 cycles for GAPDH controls in a RoboCycler (Stratagene). PCR products were analyzed on a 2% agarose gel containing 0.0025% Midori Green (Biozym). qPCR was carried out applying the Perfect CT SYBR Green master mix (Quanta) and a three-step protocol (95°C for 15 s; 57°C for 30 s; 72°C for 45 s) with an ABI 7900 HT instrument. Amplification efficiency was verified for each pair of primers using a standard curve constructed by serial dilution of a control template. Resulting C T values for AKR1B15.1, AKR1B15.2, and AKR1B10 transcripts were corrected by the average C T value of the three housekeeping genes GAPDH, HPRT, and 18S RNA (⌬C T calculated) and normalized to the expression level of AKR1B15.1 in placenta.
Activity Assays-Catalytic activity was measured using 3 H-labeled steroids either with 10 6 harvested HEK293 cells (untransfected or transfected with pIRES-hrGFP-1␣-AKR1B15) or with purified enzymes as described previously (38) with slight mod-ifications. Reaction assays using HEK293 cells contained 10 -40 nM 3 H-labeled steroid (PerkinElmer Life Sciences) and generally 350 M NAD(P)(H) cofactor (Serva, Sigma) in 500 l of reaction buffer (100 mM sodium phosphate buffer (NaP i ), pH 7.4, 1 mM EDTA). For the determination of Michaelis-Menten parameters, 0 -10 M unlabeled steroid (Sigma) was added. In assay mixtures containing purified proteins (up to 55 nM), 0.05% (m/v) BSA was added. The reaction mixtures were incubated with continuous shaking at 37°C, and the reaction was terminated by the addition of 20% stop solution (0.5 M ascorbic acid, 1% (v/v) acetic acid in methanol). Steroids were extracted on StrataC18-E (55 m, 70 Å, 100 mg/ml) SPE cartridges (Phenomenex) with methanol as eluent. Substrates and products were separated by reverse phase HPLC on a Luna 5-m C18(2) 125 ϫ 4-mm column (Phenomenex) with 43% acetonitrile in MilliQ-H 2 O as mobile phase, using a Beckman-Coulter system coupled to an online scintillation detector (Berthold LB506D). Conversion of 3 H-labeled substrate was determined from ratios of areas under the peaks by Karat software (Beckman-Coulter).
Activity assays with unlabeled substrates were carried out with purified enzyme by measuring the change in NADPH absorbance at 340 nm using a Cary50 UV-visible spectrophotometer (Varian) as described previously (2). The reaction mixture contained 150 M NADPH, 6 M purified enzyme, and substrate at variable concentrations (0 -1 mM) in reaction buffer (100 mM NaP i , pH 7.0, 1 mM EDTA). The reaction was initiated by the addition of substrate to the prewarmed mixture and allowed to run at 37°C for 10 -15 min with continuous absorbance recording. The initial velocity was calculated from the linear portion (0 -5 min) of the curve. Kinetic parameters were calculated using SigmaPlot version 12 software (Systat Software).
Fluorescence Titrations-The binding affinity of AKR1B15 isoforms to pyridine nucleotide cofactors was studied fluorometrically by following the quenching of the protein fluorescence ( ex ϭ 280 nm, em ϭ 340 nm) upon the addition of cofactors (39), using a Shimadzu RF5000 instrument. Aliquots containing 14 g of purified protein (corresponding to a final concentration of 180 or 170 M for AKR1B15.1 or AKR1B15.2, respectively) were added to 2 ml of 20 mM KP i , pH 7.4, at room temperature and equilibrated for 15-20 min. Aliquots of nucleotides were added sequentially, and changes in emission were recorded. Dissociation constants were calculated by fitting a Morrison equation (Equation 1) to the data after correction for the volume increase due to the addition of nucleotides and inner filter effect using SigmaPlot version 12 software (Systat software), where ⌬F represents decrease in protein fluorescence, E is total enzyme concentration, and N is total nucleotide concentration.
Subcellular Localization Studies-HeLa cells were grown on glass coverslips on the bottom of a 6-well plate and transiently transfected with plasmids (pcDNA3.1(ϩ), N-Myc-pcDNA3, or pcDNA4-Myc/HisB backbone) encoding AKR1B15.1 or AKR1B15.2. For counterstaining the cytoplasm and endoplasmic reticulum, pCMV-DsRed-Express2 and pDsRed2-ER (Clontech) vectors were co-transfected, respectively. After transfection, cells were incubated at 37°C and 5% CO 2 in a humidified incubator for additional 2 days. Mitochondria were counterstained before the cells were prepared for immunocytochemical analysis by incubating living cells in serum-free minimum essential medium with Earle's salts containing Mito-Tracker Orange CMTM-Ros (Molecular Probes) for 30 min. After staining the mitochondria, the cells were fixed in 3.7% formaldehyde in PBS for 10 min at culturing conditions, permeabilized for 5 min using 0.5% Triton X-100 in PBS, and blocked with 3% BSA in PBS for 1 h to prevent nonspecific binding of the antibodies. The fixed cells were consecutively incubated with mouse anti-Myc (Roche Applied Science)/goat anti-rabbit AlexaFluor 488 (Molecular Probes) antibodies in the case of Myc-tagged AKR1B15 isoforms or with rabbit anti-AKR1B15 (21st Century Biochemicals)/goat anti-rabbit Alexa-Fluor 488 (Molecular Probes) and rat-anti-AKR1B15 clone 9A5 (in-house production)/goat anti-rat AlexaFluor 488 (Molecular Probes) in the case of untagged proteins for 1-2 h. Before mounting objects on slides with VectaShield mounting medium (Vector Laboratories), nuclei were counterstained with Hoechst 33342 dye (Molecular Probes) diluted 1:5000 in PBS for 2 min. After each step, cells were washed twice with PBS. Subcellular localization data were collected and analyzed using a Zeiss AxioImager Z1/ApoTome confocal microscope with an AxioCam MRm camera and the AxioVision release 4.8 software. In silico predictions of the localization of AKR1B15 isoforms were carried out using iPSORT for non-plant proteins (40).

Two Splice Variants of AKR1B15 Are Expressed in Vivo-
Previously, we reported the identification and functional expression of a novel human member of the AKR1B family with 91% amino acid identity to the well characterized human enzyme AKR1B10 (35). The gene encoding this protein, AKR1B15, is located on chromosome 7 next to AKR1B10. After our report, a newly predicted transcript sequence corresponding to the AKR1B15 gene was deposited in the NCBI and Ensembl databases. The predicted transcript is 1621 bp in length and differs from our reported 1242-bp cDNA sequence in the 5Ј-end. Bioinformatics analysis revealed that the two sequences might result from an alternative use of the first exons of the AKR1B15 gene, leading to two transcripts, hereafter referred to as AKR1B15.1 (Ensembl transcript AKR1B15-201) for the short and AKR1B15.2 (Ensembl transcript AKR1B15-001) for the long transcript (Fig. 1A). To determine whether the alternative transcript AKR1B15.2, like AKR1B15.1, is expressed in vivo, we designed specific primers based on the predicted sequence and amplified the corresponding product by PCR from the cDNA libraries of human thymus and salivary gland. Sequencing of the resulting product confirmed its 100% identity to the sequence reported in the databases. Thus, AKR1B15.2 mRNA is expressed in tissues and might be translated into a protein because it contains an open reading frame corresponding to a 344-amino acid protein. We conclude that the AKR1B15 gene gives rise to two splice variants in vivo, presumably coding for two different protein isoforms (Fig. 1), and classify these protein isoforms as AKR1B15.1 (shorter, 316amino acid isoform, encoded by transcript AKR1B15.1) and AKR1B15.2 (longer, 344-amino acid isoform, encoded by transcript AKR1B15.2) in accordance with the guidelines for the nomenclature of alternative splicing in the AKR superfamily (41). The two AKR1B15 isoforms differ in their N termini but share the same sequence beginning with the amino acid Ser 23 in the case of AKR1B15.1 and Ser 51 in the case of AKR1B15.2 (Fig.  1B). The four amino acid residues known to comprise the catalytic tetrad in AKRs are found in both AKR variants (Asp 44 , Tyr 49 , Lys 78 , and His 111 for AKR1B15.1 and Asp 72 , Tyr 77 , Lys 106 , and His 139 for AKR1B15.2, respectively; Fig. 1B).
Tissue Distribution of AKR1B15-To determine the tissue abundance of the two AKR1B15 isoforms, we first analyzed the expression of the two mRNA splice variants in a broad panel of tissues by RT-PCR, using transcript-specific primers as depicted in Fig. 1A (sequences listed in Table 1). For compari-son, the expression of the highly homologous AKR1B10 was analyzed in the same set of samples. We found that the expression of AKR1B15.1 and AKR1B15.2 differs completely from that of AKR1B10. Whereas AKR1B10 is expressed in a fairly ubiquitous manner across the tissue panel, the AKR1B15 variants show more distinct distribution patterns ( Fig. 2A). The highest expression levels of both AKR1B15 splice variants were seen in adipose tissue, skeletal muscle, thymus, thyroid gland, and reproductive tissues (ovary, placenta, prostate, and testis). Corroborating the results from tissues, the human placental cell line BeWo and the preadipocyte cell strain SGBS expressed significant levels of both AKR1B15 transcripts ( Fig. 2A). In order to gain additional insights into the expression levels of AKR1B15, we performed qPCR on selected AKR1B15-expressing tissues, using the same transcript-specific primers as for the end point RT-PCR. We found the highest level of expression of both AKR1B15 mRNA variants in placenta, followed by adipose tissue and testes (Fig. 2B). On the absolute level, the abundance of AKR1B15 mRNA was quite low. In placenta, the tissue with Alternative splicing of the 5Ј-region of AKR1B15 leads to a totally different and longer N terminus of AKR1B15.2 compared with that of AKR1B15.1. AKR1B15.1 and AKR1B10 possess high homology in their N termini. Amino acids of the catalytic tetrad are highlighted in boldface type; the serine at position 8 of AKR1B15.1, which is mutated in a phenotype with a mitochondrial disease, is colored in blue; and the proline at position 24 of AKR1B10, which is responsible for its cytosolic localization, is colored in green. The recognition region for the monoclonal rat-anti-AKR1B15 antibody is highlighted by a green box; the C-terminal recognition region for the polyclonal rabbit anti-AKR1B15 antibody is highlighted by a red box.
the highest level of expression, the abundance of AKR1B15.1 transcripts was 100 -150-fold lower than that of GAPDH. In a majority of tissues, the AKR1B15.1 transcript was more abun-dant (at least 3-fold) than AKR1B15.2. In the thymus, prostate, and uterus comparable levels of both transcripts were found. Skeletal muscle was the only tissue where AKR1B15.2 showed

TABLE 1 Primer names and sequences used for cloning and semiquantitative RT-PCR or qPCR
Restriction sites are underlined, and coding sequences are shown in capital letters. RT, semiquantitative RT-PCR; Q, qPCR.
Recombinant Expression and Purification of AKR1B15-To verify that both AKR1B15 mRNA variants produce functional proteins, we cloned their coding regions into the vector pET28a(ϩ) and expressed encoded His-tagged proteins in E. coli. After induction with isopropyl 1-thio-␤-D-galactopyranoside, bands with the predicted molecular weights were detected in bacterial extracts transformed with the corresponding constructs (Fig. 3A). AKR1B10, which differs only in 27 amino acid residues from AKR1B15.1 (Fig. 1B), was also expressed for comparison. In contrast with AKR1B10, which is expressed as soluble protein in E. coli, both AKR1B15 isoforms were found in the insoluble fraction (Fig. 3B). We were able to solubilize both N-terminally histidine-tagged AKR1B15 isoforms using a Sarkosyl-Triton buffer system and purified both proteins to apparent homogeneity using a one-step immobilized metal (Ni 2ϩ ) affinity chromatography (Fig. 3C). We attributed the minor low molecular weight bands to the degradation or truncation products of AKR1B15 isoforms (Fig. 3C). The purification yielded 6 -10 mg of protein/liter of bacterial culture.
Nucleotide Binding-To determine the affinities of the two AKR1B15 proteins for nucleotide cofactors, we determined dissociation constants (K d ) of AKR1B15.1 and AKR1B15.2 for the four major pyridine nucleotides NADPH, NADH, NADP ϩ , and NAD ϩ using fluorometric titrations. As shown in Fig. 4A, the addition of incremental concentrations of NADPH or NADP ϩ to AKR1B15.1 led to a gradual decrease in protein fluorescence. The maximal degree of fluorescence quenching was 25-26% in both cases, and the K d values calculated from the concentration dependence of the decrease in fluorescence were 59.3 Ϯ 1.9 nM for NADPH and 60.4 Ϯ 3.5 nM for NADP ϩ . In contrast, less than 3% quenching of AKR1B15.1 fluorescence was observed with NADH and NAD ϩ in concentrations of up to 40 M after correction for inner filter effect ( Fig. 4B; titration curve not shown). These results indicate that, like other AKRs, AKR1B15.1 binds pyridine dinucleotides with high affinity and that it strongly discriminates between phosphorylated and non-phosphorylated nucleotides. The addition of any of the four nucleotides to AKR1B15.2 failed to produce any change in protein fluorescence, in agreement with the lack of detectable enzymatic activity of AKR1B15.2 (see below).
Enzymatic Activity of AKR1B15 Isoforms-We performed detailed kinetic characterizations of both AKR1B15 isoforms, AKR1B15.1 and AKR1B15.2, using a variety of physiological substrates.
Because AKR1B15 is abundant in reproductive organs (the first full-length AKR1B15 transcript was initially found in testis), we reasoned that the protein might possess enzymatic activity with sex steroids. Therefore, we tested estrogens, androgens, progesterone, and corticosteroids as potential substrates of AKR1B15.1 or AKR1B15.2. In activity assays using HEK293 cells, transiently transfected with pIRES-hrGFP1␣-AKR1B15.1 or pIRES-hrGFP1␣-AKR1B15.2 and NADP(H) or NAD(H) as cofactor, we found that neither AKR1B15.1 nor AKR1B15.2 was able to reduce or oxidize progesterone and corticosteroids (data not shown). However, AKR1B15.1 cata-  lyzed oxidation-reduction reactions with androgens and estrogens, as shown for the estrone and 17␤-estradiol pair (Fig. 5A). Catalysis was supported by NADPH or NADP ϩ but not by NADH or NAD ϩ (up to a concentration of 1.5 mM), which agrees with the binding studies, suggesting that, similar to a majority of other AKRs, AKR1B15 exhibits a strong preference for phosphorylated cofactors. In addition, we found that AKR1B15.1 possesses high positional selectivity because it catalyzed only reactions on the C17(␤) position (C17) but not on the C3 position (C3) of the steroid nucleus (Fig. 5B). In time course experiments using purified AKR1B15.1 and estrogens as well as androgens in a final concentration of 20 nM, we found that AKR1B15.1 prefers reductive over oxidative reactions and androgens over estrogens (Fig. 5C). In contrast to AKR1B15.1, AKR1B15.2 did not exhibit any enzymatic activity with estrogens, androgens, or other steroids tested. Because neither the activity assays using solubilized purified AKR1B15.2 nor the assays using HEK293 cells, in which AKR1B15.2 was expressed under physiological conditions, showed any enzymatic activity, it appears that the protein is catalytically inactive, and the lack of activity of the protein purified from bacteria could not be attributed to improper folding.
Because we found that AKR1B15.1 co-localizes with mitochondria (see below), we tested whether the enzyme displays catalytic activity with mitochondrial carbonyls or alcohols, such as acetoacetyl-CoA, oxaloacetic acid, 2-oxobutyric acid, methylmalonyl-CoA, succinyl-CoA, DL-3-hydroxybutyryl-CoA, and DL-3-hydroxy-3-methylglutaryl-CoA. We found that AKR1B15.1, but not AKR1B15.2, possessed enzymatic activity with acetoacetyl-CoA, whereas all other compounds were not detectably reduced or oxidized by either isoform (data not shown). These results indicate that AKR1B15.1 acts only on hydroxyl or keto groups of substrates possessing a bulky ring system like the steroid nucleus or the CoA.
We also determined kinetic parameters of AKR1B15.1 with different substrate classes using the purified enzyme ( Table 2). The K m values for oxidized steroids carrying a keto group on  Generation of AKR1B15-specific Antibodies and Western Blot Analysis-To examine the expression and the subcellular localization of AKR1B15 proteins, we first generated a specific polyclonal antibody that recognizes both AKR1B15 isoforms and distinguishes them from other AKR family members. Although the amino acid sequence of AKR1B15.1 is 91% identical to that of AKR1B10, there is a single stretch of six consecutive amino acids at the C terminus of the proteins (amino acids 299 -304) that is different between the two proteins. Using a peptide corresponding to this area (AKR1B15.1, amino acids 295-307; AKR1B15.2, amino acids 323-335; Fig. 1B), we were able to generate a polyclonal antibody that recognized both AKR1B15.1 and AKR1B15.2 but did not cross-react with other recombinant human AKR proteins (Fig. 6A). This polyclonal antibody did not cross-react with AKR1B10 even when loading high amounts (200 ng) of purified protein (data not shown). In West-FIGURE 6. Generation of specific polyclonal and monoclonal anti-AKR1B15 antibodies and detection of endogenous AKR1B15 isoforms. A, different His-tagged human AKRs (AKR6A3, AKR1B1, AKR1B10, AKR1B15.1, AKR1B15.2, and AKR1A1) were expressed in E. coli BL21 (DE3) and analyzed by Western blot using polyclonal rabbit anti-AKR1B15, monoclonal rat anti-AKR1B15 (9A5), or polyclonal rabbit anti-His tag (Cell Signaling) antibodies as primary antibodies and HRP-conjugated goat anti-rabbit (Invitrogen) and mouse anti-rat IgG2A (in-house production) secondary antibodies. Nontransformed E. coli BL21 (DE3) served as negative control. The anti-His tag staining showed that all proteins were expressed (bottom). B, untagged AKR1B15.1 and AKR1B15.2 were overexpressed in HEK293 cells and analyzed by Western blot using polyclonal rabbit anti-AKR1B15 and monoclonal rat anti-AKR1B15 (9A5) antibodies as primary antibodies and goat anti-rabbit (Invitrogen) and mouse anti-rat IgG2A (in-house production) antibodies as HRP-conjugated secondary antibodies, respectively. Non-transfected HEK293 cells served as negative control. Whereas the polyclonal antibody is fairly nonspecific, because it recognizes several proteins, the monoclonal antibody shows high specificity to both AKR1B15 isoforms with no cross-reactivity. C, fractions (800 ϫ g and 9000 ϫ g pellets) of the enrichment of mitochondria from BeWo cells were analyzed for the presence of endogenous AKR1B15 isoforms by Western blot using monoclonal rat anti-AKR1B15 (9A5) antibody as primary and IR dye-conjugated goat anti-rat AlexaFluor 790 (Dianova) antibody as secondary (right panel). A mixture of extracts from HEK293 cells overexpressing untagged AKR1B15.1 or AKR1B15.2 served as a positive control, and non-transfected HEK293 cells as well as a secondary antibody-only hybridization (left panel) served as negative controls. Asterisk, endogenous AKR1B15.1; arrowhead, endogenous AKR1B15.2. ern blot analysis of HEK293 transiently transfected with different plasmids encoding either AKR1B15.1 or AKR1B15.2, we could clearly detect the different overexpressed AKR1B15 isoforms. However, the polyclonal antibody also bound nonspecifically to other proteins of the HEK293 cells, including those having a molecular weight similar to that of the AKR1B15 isoforms (Fig. 6B, left). Because the high cross-reactivity of the polyclonal antibody could complicate the analysis of the expression of the native AKR1B15 proteins and because we found that antibodies against C-terminal sequences often cross-react with several other proteins, we generated a monoclonal antibody (rat anti-AKR1B15 (9A5)) recognizing both AKR1B15 isoforms and targeting a sequence more centrally located in the protein (AKR1B15.1, amino acids 114 -138; AKR1B15.2, amino acids 142-166) with high divergence compared with AKR1B10 (Fig. 1B). Like the polyclonal antibody, the monoclonal antibody also recognized both AKR1B15.1 and AKR1B15.2 and did not cross-react with other recombinant human AKRs (Fig. 6A). In contrast with the polyclonal antibody, the monoclonal antibody displayed no cross-reactivity with proteins of the HEK293 cell background when Western blots were performed (Fig. 6B, right). To examine whether the mRNA of AKR1B15 is translated to a protein in vivo, we performed Western blots of BeWo cell extracts using the specific monoclonal antibody and an IR dye-labeled secondary antibody (goat anti-rat AlexaFluor 790), which allows for the sensitive detection of low abundance proteins. Although no endogenous AKR1B15 isoforms were detectable in total cell lysates (data not shown), we were able to detect endogenous AKR1B15.1 and AKR1B15.2 in both the 800 ϫ g and the 9000 ϫ g pellet fraction of BeWo homogenates processed for the enrichment of mitochondria (Fig. 6C). Whereas AKR1B15.2 appeared as a single protein band at the expected molecular mass of 39.5 kDa, endogenous as well as overexpressed AKR1B15.1 was present as a double band corresponding to molecular masses of 36.5 kDa (expected) and 35.5-36 kDa, which could be a proteolyzed or post-translationally modified form of the protein.

DISCUSSION
In this work, we demonstrate that a novel human gene, AKR1B15, a member of the AKR superfamily, is expressed in human tissues, with the highest level found in reproductive organs, adipose tissue, and skeletal muscle. In addition, we found that the AKR1B15 gene undergoes alternative splicing, producing two open reading frames corresponding to the protein isoforms AKR1B15.1 and AKR1B15.2. Both mRNA transcripts are expressed in vivo, although expression of both variants is limited in abundance and is not as ubiquitous as that of the highly homologous AKR1B10. For the most part, the transcript AKR1B15-201 (referred to as AKR1B15.1, encoding protein AKR1B15.1) was more abundant than AKR1B15-001 (referred to as AKR1B15.2, encoding protein AKR1B15.2) in tissues expressing both variants.
Using the monoclonal rat anti-AKR1B15 (9A5) antibody, generated in house, we were able to show that both AKR1B15 transcripts are translated into protein in vivo. Consistent with the low expression of mRNA, the abundance of the AKR1B15 protein was low as well; therefore, fractionation of subcellular components was necessary to detect the endogenous protein.
Nevertheless, we were able to identify both endogenous AKR1B15 protein isoforms in the 800 ϫ g and 9000 ϫ g pellet of homogenates of the placenta-derived BeWo cells but were unable to detect endogenous AKR1B15 isoforms in commercial total protein human tissue lysates and total lysates of BeWo cells (data not shown). It appears that additional steps, such as enrichment of subcellular components or immunoprecipitation, will be necessary to characterize the expression of AKR1B15 in tissues and overcome its low abundance. Additionally, the detection of the proteins could be confounded by post-translational modifications, which could reduce the affinity of the monoclonal antibody to the endogenous protein.
Post-translational prediction algorithms indicated that both AKR1B15 isoforms could undergo various modifications (e.g. phosphorylation, SUMOylation, or ubiquitination) at several sites, including residues of the monoclonal antibody epitope (data not shown). These modifications could shift the apparent molecular weight of the proteins, destabilize the proteins, or prevent antibody binding and thus interfere with the detection of endogenous AKR1B15 proteins. Additional research is required to investigate these possibilities.
Importantly, AKR1B15.1 showed enzymatic activity with sex steroids, both androgens and estrogens, and 3-keto-acyl-CoA thioesters, such as acetoacetyl-CoA. In contrast, AKR1B15.2 appeared to be an inactive enzyme, despite the fact that it possesses all four conserved amino acid residues of the catalytic tetrad (Asp 72 , Tyr 77 , Lys 106 , and His 139 ). These findings are corroborated by the inability of AKR1B15.2 to bind nicotinamide adenine dinucleotide cofactors, whereas AKR1B15.1 binds NADPH and NADP ϩ with an affinity in the nanomolar range. AKR1B15.1 displayed absolute specificity for phosphorylated dinucleotide cofactors because neither binding nor activity has been observed with NADH or NAD ϩ .
Currently, it is difficult to distinguish whether the lack of nucleotide binding by AKR1B15.2 is due to improper folding or to an intrinsic property of this protein. The undetectable enzymatic activity in both the artificial bacterial and the mammalian expression system supports the latter hypothesis, suggesting that the long N terminus influences the protein structure in such a way that it prevents nucleotide and/or substrate access or binding. Whereas the shorter AKR1B15 isoform, AKR1B15.1, like all other known AKR1B family members, is 316 amino acids long and shares 91% amino acid sequence identity with AKR1B10, the AKR1B15.2 isoform displays greater differences because the N terminus of AKR1B15.2 has no homology with other AKR1Bs and is 28 residues longer. In the crystal structure of AKR1B proteins, the N terminus folds into a hairpin of two ␤-sheets and creates a bottom of the (␤/␣) 8 barrel (2,42,43). The alternative N terminus in AKR1B15.2 substitutes the first 22 amino acid residues of other AKR1Bs, which might lead to a disarrangement of the bottom hairpin and the first ␤-sheet of the (␤/␣) 8 barrel. This suggests the intriguing possibility that the N terminus of AKR1B15.2 might serve as a modulatory domain, regulating access to the active site, by changing its conformation in response to protein modification, such as phosphorylation. Alternatively, the non-homologous N-terminal loop of AKR1B15.2 may perform some additional function, analogous to the N terminus of AKR6 family members (Kv␤ proteins), which forms a "ball and chain" structure involved in the regulation of ion flow kinetics of voltage-gated potassium (Kv) channels (44,45). Further investigation is needed to systematically address these possibilities. A, nuclei were stained using Hoechst 33342 dye; B, mitochondria were stained using Mito-Tracker Orange CMTM-Ros; C, Myc-tagged AKR1B15.1 or AKR1B15.2 was stained using mouse anti-Myc/goat anti-mouse-AlexaFluor 488 antibodies and untagged AKR1B15.1 was stained using polyclonal rabbit anti-AKR1B15/goat antirabbit AlexaFluor 488 antibodies or monoclonal rat anti-AKR1B15/goat anti-rat AlexaFluor 488 antibodies. The individual staining as well as the overlays (D) demonstrate that AKR1B15.2 is a cytoplasmic protein, whereas C-terminally Myc-tagged AKR1B15.1 as well as untagged AKR1B15.1 co-localize with mitochondria. Nuclei are shown in blue, mitochondria in red, AKR1B15 in green, and co-localization in yellow.
Our measurements of enzymatic parameters show that AKR1B15.1 possesses K m values in the low micromolar range for 17-keto-steroids, which are similar to the K m values of other 17␤-hydroxysteroid dehydrogenases (17␤-HSDs) (e.g. HSD17B1, HSD17B5, and HSD17B12) (46 -48). We found that the enzyme is selective toward the carbonyl group located at the C17 position on the steroid nucleus. The comparatively low k cat values observed with purified enzyme may result from the purification procedure because AKR1B15.1 expressed in E. coli is an insoluble protein that needs to be reconstituted from inclusion bodies. In our k cat calculations, we assumed that all protein molecules are properly folded; however, it is more likely that only a fraction of the purified enzyme is in the right conformation. Due to the high homology between AKR1B15.1 and AKR1B10 and the fact that estrogens and androgens are substrates of AKR1B15.1, we measured the activity of purified AKR1B10 with estrone, 17␤-estradiol, ⌬4-androstenedione, and testosterone. Although it has been published that AKR1B10 is inhibited by steroids (49), we found that AKR1B10 is able to catalyze oxidation or reduction of those steroids in the nanomolar range too. However, in contrast to AKR1B15, AKR1B10 preferentially catalyzed oxidative reactions of steroids (data not shown).
Our studies revealed that AKR1B15.1 is a predominantly reductive enzyme and that it co-localizes with mitochondria. The subcellular localization was surprising because most other human AKRs are cytosolic enzymes. Although in silico subcellular localization prediction is hypothetical and often does not agree with in vivo localization (50), iPSORT predicted a mitochondrial localization of AKR1B15.1 as well as a cytosolic localization of AKR1B15.2 and AKR1B10, which is in agreement with our results. The different behavior of AKR1B15.1 and AKR1B10 concerning localization can probably be explained by the different amino acid composition in their N termini at positions 22 and 24. AKR1B15.1 possesses an arginine at position 22, and AKR1B10 features a lysine at this position, both of which have similar physicochemical properties. In contrast, the physicochemical properties of the amino acids at position 24 of AKR1B15.1 (Leu 24 ) and AKR1B10 (Pro 24 ) are clearly different. Several previous studies have shown that proline residues serve as a helix breaker (51,52). We therefore presume that Pro 24 may be the amino acid responsible for different localization of AKR1B15.1 and AKR1B10. In silico prediction with iPSORT indicated a mitochondrial location of the AKR1B10 P24L mutant. This was confirmed in localization studies with N-terminal sequences fused to GFP, in which we were able to show that the substitution P24L in the N terminus of AKR1B10 was sufficient to switch the subcellular localization of the respective GFP reporter construct from cytosolic to mitochondrial (data not shown). The results of Western blotting were consistent with the mitochondrial localization of AKR1B15.1 because endogenous AKR1B15.1 was detected in the 800 ϫ g and 9000 ϫ g pellets of BeWo cell homogenates. The 800 ϫ g pellet contains the nuclear fraction together with unbroken cells, cell debris, and remains of the supernatant, whereas the 9000 ϫ g pellet represents the mitochondrial fraction (37). Surprisingly, AKR1B15.2, which seems to be localized to the cytosol by immunohistochemistry, was found mainly in the 9000 ϫ g pellet rather than the supernatant, suggesting strong association with subcellular organelles, possibly lysosomes, which are likely to be found in the 9000 ϫ g pellet (37).
Among the AKRs, only AKR7A2 has been suggested to be associated with mitochondria in SH-SY5Y neuroblastoma cells (53). However, its rat ortholog, AKR7A4, has also been reported to be localized to the Golgi apparatus (54); therefore, the subcellular localization of this enzyme is still unclear. Hence, we conclude that we characterized the first AKR1 family member co-localizing with mitochondria; however, it still needs to be clarified whether AKR1B15.1 is located inside the mitochondria or strongly associated with the outer membrane.
Having established that AKR1B15.1 is localized to the mitochondria, we investigated whether mitochondria-specific carbonyls are potential substrates of the enzyme. We found that a 3-keto-acyl-CoA compound, acetoacetyl-CoA, can be reduced by AKR1B15.1 with a K m of about 60 M. We presume that AKR1B15.1 possesses low oxidizing activity with DL-3-hydroxybutyryl-CoA too, although the conversion could not be detected by our assays, probably due to limitations in the sensitivity of the readout of our assay, which is based on NADPH absorption. Longer chain 3-keto-acyl-CoAs, such as 3-ketopalmitoyl-CoA, could also serve as substrates of AKR1B15.1. However, up to now we were unable to verify the proposed conversion due to limitations in substrate amounts and lack of sensitive and stable detection assays. Development of a more sensitive assay, possibly based on product detection rather than nucleotide absorbance, is necessary to confirm the reaction. The free oxo-(di)-carboxylic acids oxaloacetic acid and 2-oxobutyric acid as well as the CoA-thioesters of dicarboxylic acids methylmalonyl-CoA and succinyl-CoA do not appear to be substrates of AKR1B15.1. This indicates that only carbonyl and not carboxyl groups can be reduced by AKR1B15.1. Moreover, it seems likely that all substrates need to possess a bulky ring backbone for their orientation in the substrate binding pocket of AKR1B15.1.
Although it might seem surprising that a single enzyme reduces such unrelated compounds as steroids and keto-acyl-CoA derivatives, it appears that many 17␤-HSDs exhibit a wide substrate spectrum, which includes fatty acid derivatives, bile acids, and retinoids (55). The 17␤-HSDs belong to two genetic superfamilies: AKRs and short-chain dehydrogenases (56). To date, at least 14 types of 17␤-HSDs have been identified, among which only type 5 belongs to the AKR superfamily (AKR1Cs). No activity with 3-keto-acyl-CoAs has been reported for AKR1C enzymes; however, 17␤-HSDs of types 3, 4, 10, and 12, which belong to the short-chain dehydrogenase superfamily, possess activity with both steroids and keto-acyl-CoA conjugates (55). Therefore, we propose that although structurally a member of the AKR superfamily, AKR1B15.1 functionally is a 17␤-hydroxysteroid dehydrogenase. Among the human 17␤-HSDs, HSD17B10 is a mitochondrial enzyme and catalyzes the NAD-dependent oxidoreduction of short-chain 3-keto-acyl-CoAs, along with sex steroids, as well as bile acid isomerization and glucocorticoid and gestagen catabolism (57). This enzyme is also called SCHAD (short-chain hydroxyl-acyl-CoA dehydrogenase), and it acts primarily in oxidative direction. Defects in this enzyme lead to hyperinsulinemic hypoglycemia (58, 59), abnormal thermogenesis, and lower body weight in mice (60) as well as neural disorders such as Alzheimer and Parkinson diseases, mental retardation, and infantile neurodegeneration (61).
It could be argued that the results from the activity tests are somehow inconsistent with the subcellular localization of AKR1B15.1 because AKR1B15.1 preferably catalyzes reductive reactions, but the mitochondrial matrix has an oxidative environment, where among other reactions, ␤-oxidation of fatty acids and the very first steps of the steroid synthesis (from cholesterol to pregnenolone) take place (62). However, with the current data, the role of AKR1B15.1 in mitochondria can only be hypothesized. Different studies have shown that steroids and steroid receptors are present in mitochondria and affect their metabolism (63)(64)(65). One function of AKR1B15.1 may be the activation of the steroid signaling in mitochondria, as AKR1B15.1 catalyzes, among other reactions, the conversion of biologically low active estrone to highly active 17␤-estradiol, which binds to the mitochondrial estrogen receptor with high affinity (65). Like the nuclear genome, the mitochondrial genome contains hormone-responsive elements (e.g. the estrogen response element), regulating the expression of important ribosomal and structural proteins, as well as mitochondrially encoded proteins of the oxidative phosphorylation system (64,66,67). In addition, several studies have shown that 17␤-estradiol protects the function of mitochondria in cells by reducing the amount of reactive oxygen species and therefore prevents cells from aging (64). Thus, AKR1B15.1 might provide the active steroid hormones that are known to reduce aging in mitochondria and cells. This hypothesis is supported by a recent publication by Yashin et al. (68), correlating an SNP in AKR1B15 with longevity. Another function of AKR1B15.1 might relate to the reduction of 3-keto-acyl-CoAs, such as acetoacetyl-CoA. Reduction of 3-keto-acyl-CoAs is an important step in fatty acid synthesis although not expected in mitochondria, where the reverse process, the ␤-oxidation of fatty acids, takes place. However, several investigators have been able to show that de novo fatty acid synthesis does occur in mitochondria via the FAS II pathway (69 -71) and that components of the FAS II pathway might interact with Complex I of the respiratory chain (72).
It has been recently reported that a naturally occurring mutation in the AKR1B15 gene (leading to an S8R mutation in AKR1B15.1; Fig. 1B) is linked to an infantile mitochondrial disease characterized by severe depletion of Complex I activity (73). Interactions between AKR1B15.1 and Complex I would explain why the mutation was associated with this infantile lethal phenotype (73). Here, direct protein-protein interactions in addition to the enzymatic activity of AKR1B15.1 might be of importance.
In conclusion, AKR1B15 is a novel member of the AKR superfamily with potential roles in steroid metabolism, regulation of the mitochondrial function, and aging. Given the potential role of the enzyme in several key metabolic processes, further research is required to fully characterize its substrate specificity and mechanism as well as its role in normal physiology and the significance of genetic polymorphisms in the development of pathological conditions, such as mitochondrial disease.