17β-Hydroxysteroid Dehydrogenase Type 2 Inhibition: Discovery of Selective and Metabolically Stable Compounds Inhibiting Both the Human Enzyme and Its Murine Ortholog

Design and synthesis of a new class of inhibitors for the treatment of osteoporosis and its comparative h17β-HSD2 and m17β-HSD2 SAR study are described. 17a is the first compound to show strong inhibition of both h17β-HSD2 and m17β-HSD2, intracellular activity, metabolic stability, selectivity toward h17β-HSD1, m17β-HSD1 and estrogen receptors α and β as well as appropriate physicochemical properties for oral bioavailability. These properties make it eligible for pre-clinical animal studies, prior to human studies.


Introduction
Osteoporosis is a common, age-related disease, characterized by a systemic impairment of bone mass and microarchitecture, increasing bone fragility and risk of fractures [1]. It has been shown that the drop in 17β-estradiol (E2) and testosterone (T) levels, occurring with ageing, is the main factor driving the onset and progression of this disease [2]. 17β-Hydroxysteroid dehydrogenase type 2 (17β-HSD2) catalyzes the conversion of the highly active E2 and T into the weakly potent 17-ketosteroids estrone (E1) and Δ4-androstene-3,17-dione (Δ4-AD), respectively [3]. It is expressed in osteoblastic cells [4], therefore its inhibition can lead to the desired increase of E2 and T levels in the bone tissue and may thus be a novel approach for the treatment of osteoporosis.
In the 2,5-thiophene amide class (compounds 1a-2g), h17β-HSD2 inhibitors [13,14], a broad range of inhibitory activities was detected, depending on the substitution pattern; the most active compounds show IC 50 values around 60 nM (Table 1). Conversely, the inhibitory activity towards m17β-HSD2 was only marginally affected by these changes (inhibitory activity around 30% at 1 μM). Only compound 2a shows a more pronounced m17β-HSD2 inhibition combined with good inhibition of the human enzyme (65% at 1 μM and IC 50 = 61 nM). Unfortunately, compound 2a turned out to be metabolically very unstable, with a half-life of only 4 minutes in the human liver S9 fraction [13]. In a further screening, all the tested 2,5-thiophene amide displayed a high metabolic instability [13].
It is striking that neither the nature of substituents on ring A and C or their substitution pattern does appear to exert an effect on m17β-HSD2 inhibition, whereas it is decisive for the h17β-HSD2 one. This result suggests that the inhibitors in this class are likely to have different binding modes in the two enzyme isoforms.
Exchange of the central thiophene by a 1,3-disubstituted phenyl led to compounds 4a-4c for n = 0 and 4d-4f for n = 1, with weak inhibitory activity towards both the human and the mouse enzyme (Table 1).
In contrast, compound 3a (Table 1), bearing a 1,4-disubstituted phenyl moiety as central ring, shows moderate inhibition of both h17β-HSD2 and m17β-HSD2 and also revealed exceptional metabolic stability in the human liver S9 fraction, with a half-life time>120 minutes [13]. It was therefore taken as starting point for the design of a small library of inhibitors where the substitution pattern and the physicochemical nature of substituents on the A and C rings was varied (Fig 2). A larger number of derivatives, bearing substituents with different physicochemical properties on the A ring were prepared, according to their chemical accessibilities. Compounds 25a and 25 were also synthesized to investigate the effect of the methylene linker between the amide function and the C ring.
The synthesis of the retroamide 24a, displayed in Fig 4, follows a two-step procedure. First the commercially available 3-methoxybenzoyl chloride 24c was reacted with 4-bromo-N-methyl aniline 24d according to method A and afforded the brominated intermediate 24b with 70% yield. Subsequently, Suzuki coupling following method B afforded compound 24a in 68% yield.
Compounds 18a and 24a were not tested for inhibition of m17β-HSD2, due to their low inhibitory activity towards the human enzyme. Among the different synthesized 1,4-biphenyl  amides without methylene linker (n = 0, Table 2, compounds 5a-21a), the best h17β-HSD2 inhibitory activity and selectivity toward h17β-HSD1 was achieved for compounds 6a, 7a and   17a (Table 2, IC 50 values between 260 and 330nM, s.f. between 20 and 44), showing that a 3-OMe-group on ring C in combination with either a 3-OMe-or a 3-Me-group on ring A leads to a maximum in potency and selectivity in this series of compounds.
The presence of a methyl group in 4-position of ring A is tolerated by the h17β-HSD2 (compound 17a, IC 50 = 310nM) and increases the selectivity toward h17β-HSD1 (17a: s.f. 44; 6a: s.f. 25). In contrast, compound 16a bearing a 3,4-dimethoxy substituted A ring displays a slight decrease in h17β-HSD2 inhibitory activity if compared to the corresponding compound 7a with only one methoxy group on that ring. The rigidification of the two methoxy substituents by the synthesis of a 1,3-benzodioxole ring (compounds 22a and 23a) could not overcome the drop in potency. Compounds 6a, 7a and 17a displayed the strongest m17β-HSD2 inhibitory activity (IC 50 = 260, 290 and 140 nM, respectively).
In general, the introduction of methyl-or methoxy-groups, especially in the 3-positions of rings A and B, had a positive impact on inhibitory activity, which is similar for both 17β-HSD2 isoforms (Table 2). Therefore, inhibitors belonging to the 1,4-phenyl class are likely to bind in a conserved area common to both enzyme, in contrast to the 2,5-thiophene amide class. As 17β-HSD2 belongs to the SDR superfamily, characterized by the conserved Rossmann fold and catalytic triad [23], it is possible that these inhibitors bind in or very close to these regions. Furthermore, since compounds 6a and 17a lack of the two oxygen functions to mimic the E2 interactions with the enzyme, they are likely to bind to the active site in an alternative mode, significantly influenced by the methyl substitution patterns.
Compounds with a methylene linker 25a and 25 (Table 2), showed h17β-HSD2 inhibitory activity in the same order of magnitude as the corresponding derivative 20a lacking the methylene group, but displayed improved selectivities over h17β-HSD1. In contrast to what was observed for human 17β-HSD2, compounds 25a and 25 showed a significant difference in terms of m17β-HSD2 inhibition, indicating that the hydroxy group of compound 25 might function as H-bond donor in the interaction with the enzyme. A similar behavior was observed in the 2,5-thiophene amide class with regard to the h17β-HSD2 inhibitory activity [13]: all the tested 2,5-thiophene amides with a methylene linker achieved the highest h17β-HSD2 inhibition when substituted with an hydroxy group on the C ring. It might be therefore speculated, that the addition of the methylene linker can influence the binding mode of the inhibitors in both classes.
The 1,4-phenyl amides are likely to be competitive inhibitors, as derivatives with a similar structure were found to inhibit the enzyme following this mode of action.
The inhibitory activity of compounds 6a, 17a and 25 was also evaluated in the human mammary cell line MDA-MB-231 containing endogenous 17β-HSD2 (Table 3). The compounds were tested at 250 nM and 1250 nM, representing approximately the IC 50 observed in the cellfree assay, and its 5-fold value. As displayed in Table 3, all three compounds showed an inhibition between 60% and 67%at the lower concentration and approximately 90% inhibition at the higher concentration, indicating that the inhibitors can permeate the membrane and are able to inhibit the enzyme in a concentration dependent manner. Compounds 6a, 7a, 17a, 21a, 25a and 25 were tested for their affinity toward the ERs α and β according to described methods [24] (assay details are available in the Materials and Methods). Even when applied in a 1000-fold excess relative to E2, no inhibitor was able to displace more than 20% of the steroid from the corresponding receptor, indicating a very low binding affinity to the ERs.
The metabolic stabilities of the most active compounds 6a, 17a and 25 were evaluated using human liver microsomes (S9 fraction). In addition, compounds 5a, 14a, 20a and 25a were also tested in order to investigate whether structure modifications might exert an effect on metabolic stability ( Table 4, assay details are available in Materials and Methods). All compounds, except 25, revealed a very high stability, which was not influenced by the nature of the substituents. Compound 25, exhibiting a short half-life, bears a hydroxy group, potentially susceptible to phase ΙΙ metabolism. Interestingly, all the tested inhibitors from the 1,4-phenyl amide class, for n = 0, demonstrated high metabolic stability, which seemingly constitutes a positive feature of the whole class. As species differences for the metabolism of drugs that are not or only partly metabolized are usually small [25], sufficient metabolic stability in species other than human can be anticipated.

Conclusion
The aim of this work was the design of a compound, which should be suitable for application in both an animal model of osteoporosis and in humans. We report here the discovery of compound 17a, which is the first to show an appropriate profile for this purpose, with strong inhibition of both human and mouse 17β-HSD2 and selectivity toward the respective type 1 enzymes and the ERs. It also displayed good cellular inhibitory activity, high metabolic stability and good physicochemical parameters (MW = 345 and cLogP = 4.75) predictor for good oral bio-availability [26]. A comparative SAR study for h17β-HSD2 and m17β-HSD2 is also described for the 1,4-phenyl and the 2,5-thiophene classes of inhibitors, revealing that only compounds belonging to the first series similarly inhibit the two enzymes, probably through a similar binding mode. The species specific characterization of the thiophene and the phenyl derivatives pointed out the superiority of the latter class of inhibitors, which is able to equally inhibit the two isoenzymes and additionally displays a high metabolic stability. In vivo assays in a mouse osteoporosis model will be carried out soon and the results reported in due course in a specialized journal dealing with bone diseases.

Chemical Methods
Chemical names follow IUPAC nomenclature. Starting materials were purchased from Aldrich, Acros, Combi-Blocks or Fluorochem and were used without purification. Column chromatography was performed on silica gel (70-200 μm) and reaction progress was monitored by TLC on Alugram SIL G/UV254 (Macherey-Nagel Mass spectrometry was performed on a TSQ Quantum (ThermoFisher, Dreieich, Germany). The triple quadrupole mass spectrometer was equipped with an electrospray interface (ESI). The Surveyor-LC-system consisted of a pump, an auto sampler, and a PDA detector. The system was operated by the standard software Xcalibur. A RP C18 NUCLEODUR 100-5 (3 mm) column (Macherey-Nagel GmbH, Dühren, Germany) was used as stationary phase. All solvents were HPLC grade. In a gradient run (acetonitrile/water) the percentage of acetonitrile (containing 0.1% trifluoroacetic acid) was increased from an initial concentration of 0% at 0 min to 100% at 15 min and kept at 100% for 5 min. The injection volume was 15 μL and flow rate was set to 800 μL/min. MS analysis was carried out at a needle voltage of 3000 V and a capillary temperature of 350°C. Mass spectra were acquired in positive mode from 100 to 1000 m/ z and UV spectra were recorded at the wave length of 254 nm and in some cases at 360 nm.
All microwave irradiation experiments were carried out in a 507 CEM-Discover microwave apparatus.
Method A, general procedure for amide formation. To a solution of bromobenzoylchloride (2 mmol) was added the corresponding N-methylaniline (2 mmol) followed by Et 3 N (2 mmol) in CH 2 Cl 2 (10 mL) at 0°C. After a few minutes, the ice bath was removed and the reaction mixture was warmed up to room temperature and stirred at room temperature overnight. The reaction mixture was extracted twice with CH 2 Cl 2 (2 × 15 mL). The organic layer was dried over MgSO 4 , filtered and the solution was concentrated under reduced pressure. The residue was purified by silica gel column chromatography using hexanes and EtOAc as eluent or by trituration in a mixture of diethyl ether / petroleum ether to afford the desired compound.

logP Determination
The logP values were calculated from ACD/Labs Percepta 2012 Release program. The logarithm of partition constant P (log P) was calculated using the "GALAS" method (Global Adjusted Locally According to Similarity). The program predicts clogP by comparing the molecule with structurally similar molecules where experimental data are known.
Ethics statement. Anonymized placental samples were obtained from Saarbrücken-Dudweiler Hospital's Department of Gynecology. No author involved in this study has received information about the patients. The microsomal fraction of the mouse enzyme (m17β-HSD2) was obtained from mouse livers, which were bought from Pharmacelsus GmbH (Saarbrücken, Germany).