cis–trans Isomerization of silybins A and B

Methods were developed and optimized for the preparation of the 2,3-cis- and the 10,11-cis-isomers of silybin by the Lewis acid catalyzed (BF3∙OEt2) isomerization of silybins A (1a) and B (1b) (trans-isomers). The absolute configuration of all optically pure compounds was determined by using NMR and comparing their electronic circular dichroism data with model compounds of known absolute configurations. Mechanisms for cis–trans-isomerization of silybin are proposed and supported by quantum mechanical calculations.


Introduction
The flavonolignan silybin (alternative name silibinin), occurring in the fruits of Silybum marianum (milk thistle), consists of two stereoisomers -silybin A (1a) and B (1b) -in a ca. 1:1 ratio (Figure 1). Their absolute configuration is known [1,2] and their separation was accomplished recently [3][4][5]. Both silybin isomers as well as other flavonolignans from silymarin (crude defatted extract from the fruits of S. marianum) are products of a phenolic oxidative coupling of the flavonoid taxifolin and the lignan coniferyl alcohol. The mechanism of this coupling reaction was described [6] and implied a 10,11-trans relative con-  figuration, as in all the major components of silymarin. Little is known about the structures of the minor components of silymarin. It was speculated that some of them were 10,11-cisanalogues of the major silymarin constituents [7], but these minor components were never isolated in sufficient amount and purity to enable such unwarranted hypotheses to be verified [8,9]. Another study focusing on the minor components of silymarin identified two new compounds isosilybin C (3) and D (4), however these were shown to be regioisomers of isosilybins A (5) and B (6) [10] (Figure 2).
Other authors reported 2,3-cis-isomers of silybin [11,12], the relative configuration of which were corroborated by 1 H NMR coupling constants, i.e., J 2,3 of ca. 11 Hz in the trans-isomers and 2-3 Hz in the cis-isomers. Nevertheless, their absolute configurations remained unknown. The origin of the cis-isomers either as biosynthetic side products or as artifacts formed during their isolation is also unknown.
The isolation of naturally occurring cis-isomers in the pure form and their complete structure identification would be the only unambiguous way how to prove identity of natural and synthetic cis-isomers. Unfortunately, the conventional identification by LC-MS or UV-vis scanning is inadequate as the major and minor compounds exhibit the same MS and UV profiles.
The content of silybin cis-isomers in silymarin is presumably very low. The composition of silymarin strongly depends upon its source, which is influenced by the variety of S. marianum and by the cultivation, harvest and processing conditions. Variations of the minority content (not only silybin cis-isomers) is even more pronounced. Therefore, their identification in a particular silymarin preparation would have only limited information value.
Silybin is an important pharmaceutical commodity and a complete understanding of its composition is needed in order to understand its pharmaceutical properties. Therefore, the detailed structural knowledge and availability of (potential) minor impurities is a fundamental requisite, e.g., for master file assembly.
The aim of this work was to prepare stereochemically pure 2,3cisand 10,11-cis-isomers of silybin A (1a) and B (1b) by a Lewis acid catalyzed isomerization to determine their absolute configuration and to propose a mechanism for the cis-transisomerization processes.

Results and Discussion
Chemistry In our previous work on the enzymatic kinetic resolution of silybin [4,5], BF 3 •OEt 2 in EtOAc was found to catalyze the transesterification of silybin to yield not only 23-O-  acetylsilybin (2, ca. 90%), but two novel compounds with UV spectra similar to that of silybin. Enzymatic alcoholysis (n-butanol) of this mixture by Novozym 435 led, after a prolonged reaction time, to the removal of the acetyl groups from trans-isomers 2a and 2b to give 1a and 1b (Figure 1), while the two minor isomers remained acetylated. Separation of the resulting mixture by silica gel column chromatography yielded an inseparable mixture of the two new compounds. In HPLC they were assigned to two separate peaks with the same molecular mass (m/z 524). 1 H NMR spectra indicated that the two compounds differ only slightly from the silybin derivatives 2a and 2b, respectively. The unknown compounds exhibit a J 10,11 of ca. 3 Hz, whereas natural silybin (1) has a J 10,11 of ca. 8 Hz. Based on this evidence, the unknown compounds from the acylation reaction were proposed to be a diastereoisomeric mixture of 23-O-acetyl-10,11-cis-silybin A (7) and B (8) ( Figure 3) in a ca. 1:1 ratio.
As the yields of the two 10,11-cis-silybin isomers were low (ca. 10%), we optimized the reaction conditions for better yields. In the initial screening, aimed at finding the most suitable solvent and Lewis acid (Table 1), we used natural silybin, i.e., ca. 1:1 mixture of 1a and 1b.
The choice of solvent was limited by the low solubility of silybin in most organic solvents or by their incompatibility with the Lewis acid BF 3 ·OEt 2 . The following solvents were tested: EtOAc, DMF, CH 3 CN, DMSO, CHCl 3 at 0, 25, and 50 °C. In CH 3 CN and DMSO silybin decomposed, and in CHCl 3 the reaction failed. In EtOAc, the reaction was accompanied by C-23 O-acetylation, which complicated the reaction mixture analysis. DMF at 50 °C was found to be the most suitable solvent for the isomerization and was used for further Lewis acid screening. Eventually, BF 3 ·OEt 2 proved to be the most suitable Lewis acid for silybin isomerization (Table 1). Other Lewis acids either gave lower isomerization yields (SnCl 4 ), did not work at all, or even caused decomposition. Toluene-4sulfonic acid, as a representative protic acid, gave no reaction under the same conditions.

Scheme 2: Silybin B isomerization in EtOAc.
isomer 13 (Scheme 3). Surprisingly, the absolute configuration at C-2, C-3 of compound 13 was the opposite of 1a (2S,3S vs 2R,3R), which was confirmed by a comparison of their electronic circular dichroism (ECD) data. Formation of the 10,11cis-isomer with the 2R,3R configuration was not observed during/after the isomerization of 1a in EtOAc. All the reactions in EtOAc were accompanied by C-23 O-acetylation.

Separation and purification of 2,3-cis-silybins
The separation of 2,3-cis-silybin 7 from unreacted 2,3-transsilybin 1a (or 10 from 1b) on silica gel was not feasible, analogous to the preparative separation of silybin stereoisomers 1a and 1b (Figure 1), which was a challenge for several decades. It was accomplished as the separation of the respective silybin glycosides for the first time [13,14], later by HPLC [3], and at the preparatory scale by lipase-catalyzed discrimination [4,5]. However, chromatographic separation of 2,3-cis-and 2,3-transsilybins is feasible after their C-23 O-acetylation, which can be accomplished with Novozym 435 in an acetone/vinyl acetate mixture, giving a nearly quantitative yield of the respective acetates (without the risk of further isomerization) (Scheme 4, part I). As the 2,3-trans-isomer was more abundant than the 2,3-cis-isomer its excess was removed before enzymatic acetylation by crystallizing the crude mixture from MeOH/H 2 O 9:1, which increased the ratio of the cis/trans-isomer in the mixture from 1:4 to 1:2.

Deacetylation of cis-23-O-acetylsilybins
All pure cis-silybins were obtained as their C-23 acetates. The deacetylation proved to be rather difficult, as both acidic and basic hydrolysis as well as mild transesterification (NaOMe, MeOH) failed and/or resulted in decomposition or reversed isomerization to the original trans-isomers. We eventually had to use enzymatic alcoholysis with a large excess of Novozym 435 and a longer reaction time (5:1, wt. enzyme/substrate, 3-5 d, 45 °C) compared to the alcoholysis of the respective acetates of the trans-silybins. This method finally yielded pure 2,3-cis-silybins A (9) and B (10) as well as 10,11-cis-silybin B (14) (Scheme 4, parts II and III).

Determination of absolute configuration of cis-silybins
The determination of the absolute configuration of the new cissilybins by X-ray analysis is not possible, since all attempts to obtain suitable crystals using various crystallization conditions were unsuccessful. Pure silybins and their congeners are known for their poor crystallization or the formation of small crystals, which are not suitable for X-ray analysis. The absolute configuration of silybins A (1a) and B (1b) was determined indirectly on the basis of a comparison of their physicochemical data (NMR, ECD, OR). Later, these data were matched to those of their 10,11-regioisomer, isosilybin A (5) [11]. The same approach was used to determine the absolute configuration of isosilybin B (6). The only analogue that yielded suitable crystals, and hence an unequivocal determination of its absolute configuration, was 7-(4-bromobenzoyl)isosilybin A [15].
In this study the determination of the absolute configuration was based on NMR and ECD spectroscopy. Their relative configurations, i.e., the determination of trans/cis isomers at C-2, C-3 and C-10, C-11, are set by the 1 H NMR coupling constants J 2,3 and J 10,11 , respectively. The trans-isomers exhibit J 2,3 of ca. 11 Hz and J 10,11 of ca. 8 Hz, while the cis-configuration is indicated by a value of J 2,3 or J 10,11 of ca. 2-3 Hz [11,12].
For the assignment of the absolute configuration, experimental ECD spectra of compounds 9, 10, 13, and 14 were compared to those of related compounds with known absolute configurations [2]. The assignment of Cotton effects (CE) to particular regions in this molecule is based on the rough assumption that silybin and its analogues are considered to be composed of two separate π-conjugated subsystems, the 3-hydroxyflavonone Scheme 4: Schematic flowchart of the procedures for the preparation and the isolation of cis-silybin isomers (cis-isomers are underlined).
The ECD of the 3-hydroxyflavonone moiety was discussed with the aim to determine absolute configuration [21]. UV absorption bands in the respective 270-290 nm and 330-320 nm ranges were utilized, which were clearly rationalized in terms of the corresponding electronic transitions [22,23]. A pair of positive-negative CEs at both spectral ranges is characteristic of the (2R,3R) configuration of the 3-hydroxyflavonone moiety. Nevertheless, it should be kept in mind that there are also other absorption bands at shorter wavelengths giving rise to observable CEs near 230 nm, which are relatively constant in pattern [24][25][26][27][28].
To assign the absolute configuration at C-10 and C-11 of silybin we combined the corresponding coupling constants J 10,11 with the CEs in 200-280 nm spectral range. It was shown that a sign of the CE at ca. 236 nm could be used to determine the stereochemistry at C-3 (i.e. C-10 of silybin) (negative CE corresponds to the R configuration) [2,17,20]. Thus, the negative CE around 230 nm for 13, corroborated by the vicinal 1 H-1 H coupling constants (J 10,11 = 2.9 Hz, e.g., cis-configuration) ( Figure S4, Supporting Information File 1) indicates an absolute configuration of 2S,3S,10R,11S. In contrast to the ECD spec- trum of silybin B (1b) ( Figure S3, Supporting Information File 1), the J 10,11 coupling constant (2.8 Hz, cis-configuration) of compound 14 and its ECD spectrum (positive/negative CE around 240 nm) implies the absolute configuration at C-10, C-11 to be 10S,11R, so that the absolute configuration of 14 is 2R,3R,10S,11R.

Mechanism of silybin isomerization
Based on the absolute configuration of the new cis-silybin isomers, we propose mechanisms for the stereospecific isomerization.
The isomerization process is initiated by BF 3 complexation. Boron trifluoride may complex silybin through a coordinate bond, in which the two electrons originate from the oxygen atoms of silybin. Silybin and BF 3 are hard bases (η = 4.8 eV in EtOAc) and acids (η = 7.3 eV in EtOAc), respectively (Table 2), favoring such a coordinate bond. The hardness of BF 3 and silybin A is not significantly modified when the solvent polarity is increased ( Table 2). The hardness of silybin B was not calculated, as the stereochemistry is not expected to significantly modify this parameter. The hardness calculation is based on the HOMO (highest occupied molecular orbital) and LUMO (lowest unoccupied molecular orbital) energies, which are the same for silybin A and B. The quantum calculations show that the coordination complex at C-4=O (IIa in Scheme 5) is the most stable ( Table 3). The corresponding complex IIa exhibits a strong O-B bond of 1.55 Å, lower than, e.g., 1.73 Å in the complex at O-1 (Table 4). In the benzopyranone moiety, IIa is more stable than the complex at O-1 by ca. 6 kcal·mol −1 .
In principle, the isomerization can be initiated by the formation of coordination complexes at either C-4=O or O-1 (see Scheme 5). Nevertheless, based on the quantum chemical calculations (Table 3 and Table 4) together with the representation of cis-isomers in the isomerization mixture, the most stable coordination complex IIa at C-4=O is stabilized by a strong O-B bond of 1.55 Å, having an atomic charge at C-3 close to zero (Table 4). This favors a proton release from this position and the formation of the negatively charged species IIb. The release of BF 3 probably occurs from IIb by two different routes. The major pathway (route I) proceeds without ring opening and after re-protonation it leads to the major cis-isomer 12. The minor route (route II), accompanied by a ring opening of the benzopyranone, proceeds via intermediate IIc, which after ring closure and protonation yielded a mixture of 2a and 13. This reversible benzopyranone ring opening/closure process is probably responsible for the inversion of the absolute configuration at C-2, C-3 of 13 compared to 2a (Scheme 5).
The limited degree of isomerization at C-2 of silybin requiring the ring opening of the benzopyranone is in agreement with a related study of taxifolin isomerization [34]. A deuterium incorporation NMR study supported by quantum chemical calculations showed that the intermediate of taxifolin with an open benzopyranone ring (α-hydroxychalcone related to IIc in our study) is very short-lived and the energy barrier for its formation is relatively high [35]. It must be noted that IIc contains additional substitution at the ortho-dihydroxyphenol moiety, which probably further decreases the possibilities for resonance stabilization compared to unsubstituted α-hydroxychalcone (quinone methide formation is disabled). Interestingly, the course of 1a and 1b isomerization was slightly different in DMF than in EtOAc, since it only yielded the cisisomers 9 and 10. The isomer with an inverted absolute configuration at C-2, C-3 (the non-acetylated form of 13) was not observed at all. This is easily explained by the charge distribution of complex IIa. In DMF, the atomic charge at C-3 is closer to zero than in EtOAc (Table 4). This clearly indicates that the formation of the open intermediates (IIc) is not supported in DMF.
The isomerization of 1 at the benzodioxane moiety is also initiated by BF 3 complexation, followed by a multistep mechanism including the ring opening of benzodioxane. As the change in configuration takes place exclusively at C-11, it is probably initiated by the acid-catalyzed dioxane ring opening at O-12 (Scheme 6). This site is highly nucleophilic in EtOAc (Figure 7a), with lone electron pairs, which is in favor of BF 3 complexation as confirmed by the complexation energies (Table 3). Following complexation, the D-ring opening proceeds (Scheme 6), allowing molecular rearrangements and isomerization after ring closure.
The observation of the isomerization of 1 (or 2) at C-11 that occurs in EtOAc but not in DMF can be explained either by a solvent effect or by the participation of the acetyl group. However, the second possibility is less probable, since the participation of the acetyl group should lead to the formation of an isomer with a retained configuration at C-11 (two subsequent nucleophilic substitutions with two Walden inversions). The possibility of acetyl group participation was refuted by the reaction of 2 in DMF, so that no isomerization took place at C-11. Moreover, in DMF the f(r) Fukui function is totally displaced into the E-ring (Figure 7b), so that the O-12 atom loses its nucleophilic character, which partially rationalizes the solvent effect observed here.

Conclusion
We report here the first syntheses of hitherto undescribed cisisomers of silybin in optically pure form starting from silybin A (1a) and B (1b). The determination of their absolute configuration was based on an analysis of NMR coupling constants and a comparison of their ECD spectra with model compounds with well-defined absolute configurations. Moreover, the absolute configuration of these novel 2,3-cis-and 10,11-cis-isomers of silybin enabled us to propose mechanisms for the cis-trans isomerization of silybin. Although analogous isomerizations of similar compounds at respective C-2, C-3 centers have been described, we present here parallel isomerizations on both chiral centers under kinetic control. cis-Silybins were obtained by the chemoenzymatic separation methods mostly due to the sensitivity of new cis-derivatives to the reversed isomerization.

Experimental
General experimental procedures NMR spectra were recorded in DMSO-d 6 (30 °C) by using a Bruker AVANCE 600 NMR spectrometer (600 MHz for 1 H, 151 MHz for 13 C) with the residual solvent peak as the internal standard. Mass spectra were measured in an APEX-Ultra FTMS with ESI ionization. The high-resolution ESI-MS spectra were measured by using a GCT Premier benchtop orthogonal acceleration time-of-flight mass spectrometer.
[α] D 22 −51.6 (c 0.091, acetone); ECD spectrum, see Supporting Information File 1, Figure S2; HRMS (100 mL) and the mixture was kept for 48 h at 80 °C. The reaction mixture was quenched by the addition of an ice-cold solution of saturated NaHCO 3 (100 mL), and after stirring for 10 min both phases were separated. The aqueous phase was extracted with EtOAc (3 × 50 mL). The combined organic layers were dried (Na 2 SO 4 ) and evaporated to dryness. The crude mixture was purified by column chromatography (CHCl 3 / acetone/toluene/HCO 2 H, 95:5:5:1, twice) yielding a mixture of 2b and 8, which was then dissolved in a mixture of MTBE/nbutanol (150 mL, 9:1 v/v), Novozym 435 (0.25 g, ≥10000 U/g, 100% w/w) was added, and the mixture was shaken at 45 °C and 650 rpm for 37 h until the ratio of 8/2b was 96:4 (HPLC). After enzyme removal by filtration, the solution was evaporated, and the crude mixture purified by column chromatog-  Figure S4; 1 H and 13 C NMR data, see Table 7 Figure S3; 1 H and 13 C NMR data, see

Calculation methods
All geometries and energies, including the zero-point correction (V), enthalpies (H) and Gibbs energies (G) at 298 K of the reactants, intermediates and products were determined at the (U)B3P86/6-31+G(d,p) level, well-adapted for polyphenol reactivity. Solvent effects were implicitly taken into account by using a PCM (polarizable continuum model) method; the IEFPCM (integral equation formalism PCM) method coupled to UA0 radii was used. The mechanisms were studied with both EtOAc and DMF solvents with a dielectric constant ε of 5.99 and 37.21, respectively.
The hardness η is related to the hard-soft-acid-basis (HSAB) principle. According to this theory, hard acids react with hard bases whereas soft acids react with soft bases. The hardness is given by where I and A are the adiabatic ionization potential and the adiabatic electron affinity, respectively. The hardness of silybin B was not calculated, as the stereochemistry is not expected to significantly modify this parameter. The hardness calculation is based on the HOMO (highest occupied molecular orbiral) and LUMO (lowest occupied molecular orbital) energies, which are the identical for silybin A and B. The hardness was calculated in benzene, EtOAc, and DMF to study the impact of the solvent polarity on the HSAB principle. The hardness of BF 3 and silybin A is not significantly modified when the solvent polarity is increased (Table 2).
New methods to rationalize chemical reactivity have been developed in the field of quantum mechanical methods over the past few years. The Fukui function f k (r) has become one of these powerful tools, providing an atomic picture of hardness. For a given atom k, it is given by where f k + (r) and f k − (r) are the electrophilic and nucleophilic contributions of the Fukui function calculated as follows: and , where q k (N), q k (N−1) and q k (N + 1) are the electronic population of atom k in its neutral, radical-cation and radical-anion forms, respectively. In this study, the Fukui function is used to partially rationalize BF 3 complexation. In this case, the nucleophilic contribution is the most important parameter. It must be stressed that the higher the f k − (r), the higher the atomic nucleophilic capacity. All calculations were carried out by using the Gaussian09 software [35].

Supporting Information
Supporting