Scalable enantioselective total synthesis of taxanes.

Taxanes form a large family of terpenes comprising over 350 members, the most famous of which is Taxol (paclitaxel), a billion-dollar anticancer drug. Here, we describe the first practical and scalable synthetic entry to these natural products via a concise preparation of (+)-taxa-4(5),11(12)-dien-2-one, which has a suitable functional handle with which to access more oxidized members of its family. This route enables a gram-scale preparation of the 'parent' taxane--taxadiene--which is the largest quantity of this naturally occurring terpene ever isolated or prepared in pure form. The characteristic 6-8-6 tricyclic system of the taxane family, containing a bridgehead alkene, is forged via a vicinal difunctionalization/Diels-Alder strategy. Asymmetry is introduced by means of an enantioselective conjugate addition that forms an all-carbon quaternary centre, from which all other stereocentres are fixed through substrate control. This study lays a critical foundation for a planned access to minimally oxidized taxane analogues and a scalable laboratory preparation of Taxol itself.

Introduction

Terpenes are omnipresent natural products mainly found as constituents of plant oils, and have long held importance as flavours and fragrances, and as poisons and medicines.[1-3] Isolation, structural elucidation and synthetic studies on this vast family of natural products were already being conducted over 100 years ago,[1] providing constant challenges to chemists and thus resulting in a continuous development of the field. Taxol® (1), a successful anti-cancer drug, is one example of societal and scientific merit, being one of the most famous terpenes from a medicinal standpoint and one of the most densely functionalised and complex molecules from a structural standpoint. Its unique mechanism of action involving the stabilisation of microtubules[4] has been fascinating biologists as well, and thus extensive research efforts have been carried out around the globe by chemists and biologists alike.

Taxol® (1) is a potent anticancer drug originally used as a treatment for ovarian and breast cancer,[5] but its versatile effects now extend to treatments of lung, liver and other types of cancer.[6] Initially isolated from the bark of the yew tree,[7] substantial work in the 1980s and 1990s allowed for a semisynthetic route to Taxol® (1) that reduced the accompanying detrimental effects to the yew tree population,[8-10] and currently, its commercial supply relies on cell-culture production (Taxol® is currently manufactured with plant cell-culture technology through a collaboration of Bristol-Myers Squibb and Phyton Biotech, Inc., a DFB Pharmaceuticals company). This diterpene is one of few natural products that continues to engender worldwide attention, and tremendous efforts have already culminated in seven total syntheses[11-19] and one formal synthesis.[20] However, the inventory of man-made Taxol® (1) produced by total synthesis is still less than 30 mg,[11-20] even when including the most recent approach by Takahashi wherein an automated synthesis was employed for the majority of their synthetic steps.[20] In contrast, the current industrial output of 1 through the use of biological machinery is of ton-scale (109 mg), clearly indicating the magnitudes of difference in chemical versus biological scale and efficiency. Closing this huge gap is not simply a matter of engineering, but rather a learning objective with which to guide chemical innovation, in the tactics, strategies,[21] and methods of synthesis(for a discussion of the terms ”strategy” versus ”tactics”, see reference 21). Up until the pioneering bioengineering work of Stephanopoulos in 2010, there was no access to large quantities of the less-oxidised taxane members such as taxadiene (7).[22] Although not necessarily as a means of supplanting its current ton-scale production, a reexamination of Taxol® (1) over 15 years after its first total syntheses could be of academic and medicinal merit. In this article, we present the first scalable synthetic entry to the taxane family, setting the stage for rapid access to minimally oxidised taxane analogs and a scalable laboratory preparation of Taxol® (1) itself.

Inspired by nature’s efficiency in creating vast numbers of complex terpene natural products, a research program was initiated in 2007[23] to mimic the essence of the two-phase biosynthesis of terpenes.[24] In the first biogenetic phase (the cyclase phase), linear hydrocarbon building blocks are brought together and cyclised efficiently, and in the second phase (the oxidase phase), C=C and C–H bonds are oxidised in a divergent manner, thus generating structural diversity. The taxanes represent a large family of terpenes comprising over 350 natural products,[25-32] for which this two-phase terpene synthesis strategy, if realised in a laboratory setting, could target not only Taxol® (1) but also related taxanes that differ in oxidation levels (Figure 1).[23,33] As such, this research program is designed to recapitulate the way that nature builds taxanes. The goal is to divergently access all ”pre-taxol” compounds (both natural and unnatural) and to unveil new insights to chemical reactivity during an ”oxidative ascent” of the taxane pyramid. This article outlines the completion of our artificial ”cyclase phase ”for the taxane terpene family, as a prelude to a total synthesis of Taxol® (1). Thus, the logic (including strategic and tactical concerns)[21] and execution of a concise and scalable synthesis of minimally oxidised taxanes, including nature’s putative cyclase phase endpoint, taxadiene (7),[34] are reported.

Figure 1

Retrosynthetic analysis of Taxol® (1) and other members of the taxane family

a, Partial ”oxidase phase pyramid” for the retrosynthetic planning of the taxane family, including its key member, Taxol® (1). b, Representative taxanes of varying oxidation states, sharing a C2-hydroxyl group. c, Synthetic design for ”taxadienone” (6) and reduction to generate taxadiene (7). Note: Sites of oxidation installed onto taxadiene (7) are indicated in red. The ”oxidation level” of taxanes is defined as the number of C=C and C–O bonds installed onto the taxane carbon skeleton.[33]

Structurally, Taxol® (1) is a highly oxygenated diterpene adorned with acetyl and benzoyl groups, as well as a signature taxol side chain at the C13 oxygen atom (see carbon numbering on 1, Figure 1a). For retrosynthetic discussion purposes, 1 is treated as if it were devoid of acyl groups, and this target is substituted with oxygenated hydrocarbon 2.[33] Placing 2 at the apex of an ”oxidase phase pyramid”, the number of C–O bonds decreases as one moves down the pyramid, with taxanes 3, 4 and 5 corresponding to the oxidation pattern of 2-acetoxybrevifoliol,[25] taxinines E and J,[26] and yunnanxane,[27] respectively (Figure 1b). These highly oxidised taxanes all feature a C2-hydroxyl group and can be further simplified to taxa-4(5),11(12)-dien-2-one (6), dubbed ”taxadienone”, which represents a benchmark intermediate for a comprehensive access to the taxane family. Furthermore, if taxadiene (7) were to be desired, one could simply deoxygenate 6 to generate the least oxidised natural product in the taxane family (Figure 1c). A full scholarly analysis of the taxane pyramid has been previously described, along with the merits in choosing a C2-oxidised taxane such as 6.[33]

Strategic disconnections of 6 were built upon the landmark syntheses of 1[11-20] and the many studies geared toward the taxane framework.[35-41] Out of the many possible disconnections of the 6-8-6 tricyclic backbone, one was chosen such that: 1) the forward synthesis would be short, convergent, and scalable;[42] and 2) its asymmetric synthesis would only rely on one enantioselective reaction, after which the resulting stereochemical information could be propagated to set all other stereocentres diastereoselectively. A vicinal difunctionalisation/Diels–Alder strategy to forge the taxane AB ring system (see ring numbering in 2, Figure 1) seemed to adhere most to the above criteria. The Diels–Alder strategy was previously utilised in the context of taxane synthesis by Shea,[35] Jenkins,[36] Williams[37] and others,[38-41] and in fact, Williams has reported the only total synthesis of (±)-taxadiene (7), in 26 steps, back in 1995.[37] Furthermore, the decision to implement a vicinal difunctionalisation strategy was underpinned by recent developments in the asymmetric formation of all-carbon quaternary centres.[43,44] Thus, a concise, enantioselective strategy was conceived, allowing for a gram-scale access to 6 in only seven steps, as well as a gram-scale synthesis of (+)-taxadiene (7).

Results

In the forward direction, known compounds 10[35,40,41] (made from 8 in a modified one-pot procedure; see Supplementary Information) and 11[45] (made from 9 using a known[45] one-step procedure) were combined to generate enone 12 by means of a Lewis acid-modified organocopper 1,6-addition (Figure 2). This convergent synthesis of 12 is operationally simple and provides decagram quantities per reaction batch (86%; over 50 grams of this material has been synthesised to date). This enone was deemed to be a suitable substrate for Alexakis’ enantioselective conjugate addition[43] to establish the quaternary centre (C8). However, this strategy was contingent upon the feasibility of trapping the resulting aluminum enolate by TMSCl. Although Alexakis noted that isolation of such silyl enol ethers is difficult due to facile desilylation,[43] a modified quenching procedure involving dilution with THF, addition of TMSCl, followed by addition onto Florisil® in 1:10 Et3N/hexanes, allowed the isolation of TMS enol ether 14 in 89% yield and in 93% ee (1.0 g scale; enantioselectivity was measured for desilylated ketone 15 on chiral HPLC; see Supplementary Information). Only 2 mol% of CuTC and 4 mol% of chiral phosphoramidite ligand 13 were required,[43] and this asymmetric reaction was routinely conducted on gram scale (over 20 grams of this material has been synthesised to date). TMS enol ether 14 was indeed quite prone to desilylation, and formation of undesired ketone 15 was often the sole reaction pathway when attempting Mukaiyama aldol reactions[46] with commonly used Lewis acids (such as TiCl4, SnCl4, Sn(OTf)2, Sc(OTf)3, BF3•OEt2, TMSOTf, ZnCl2, or MgBr2•OEt2) or other catalysts (such as LiClO4, Zr(OtBu)4, Bi(OTf)3, AgOTf, or SiCl4), even when performed under strictly anhydrous conditions. Surprisingly, water was found to be the enabling additive that allowed for an aldol reaction with acrolein. Gratifyingly, Kobayashi conditions[47] with Gd(OTf)3 in 1:10:4 H2O/EtOH/PhMe generated the corresponding aldol product smoothly as a 2:1 mixture of diastereomers that are isomeric at C3. This aldol product was then oxidised in the same reaction pot with Jones’ reagent to generate keto-enone 16 in 85% over two steps (3.2 g scale, 2:1 ratio of diastereomers at C3; this inseparable mixture of stereoisomers was carried onward for the ensuing step). A Diels–Alder reaction to generate the 6-8-6 tricyclic skeleton has been previously demonstrated on similar systems,[35-37] and the mixture 16 reacted predictably with BF3•OEt2 to furnish tricyclic compound 17 (2.3 g scale). The desired diketone 17 was obtained in 47% yield, along with its diastereomer (see Supplementary Information) in 29% yield, with complete diastereoselectivity at C1. These cyclised diketones displayed very different polarities and were thus easily separated by silica gel chromatography at this stage. Only one more carbon remained to be installed to complete the taxane framework: this was achieved via enol triflate formation followed by Negishi coupling to generate taxadienone (6) in 84% over two steps (2.4g scale). This material solidified quite readily, and a single-crystal X-ray analysis confirmed both the absolute and relative stereochemistry of the synthesised ABC ring system. This lowly oxidised core is the key intermediate for our future research efforts to elaborate the taxane pyramid. In summary, the total synthesis of (+)-taxadienone (6) was achieved in a total of 8 steps, with a longest linear sequence of 7 steps. A few grams of taxadienone (6) could be synthesised by one chemist over the course of 7 days, with an overall yield of 18% from 8 or 20% from 9.

Figure 2

Enantioselective synthesis of key taxane 6

Conditions: (a) 2,3-dimethyl-2-butene, CHBr3, potassium tert-butoxide, hexanes, 2 h; evaporate volatile materials, then N,N-dimethylaniline, 150 °C, 30 min (67 %); (a’) 3-ethoxy-2-cyclohexen-1-one, vinylmagnesium bromide, Et2O, 16 h (75%);[45] (b) 10, sec-butyllithium, Et2O, −78 °C, 15 min; then CuBr•SMe2, 30 min; then TMSCl, 5 min; then 11, 2 h; warm tort, 8 h; then AcOH, 30 min; then 3N HCl, 30 min (86%); (c) CuTC (2 mol%), phosphoramidite 13 (4 mol%), Et2O, rt, 30 min; then 2.0 M Me3Al, enone 12, −30 °C, 24 h; then THF, TMSCl, 0 °C to rt, 8 h; then Et3N, Florisil®, 2 h (89%, 93% ee); (d) Gd(OTf)3 (10 mol%), acrolein, 1:10:4 H2O/EtOH/PhMe, 4 °C, 24 h; then evaporate volatiles, then Jones’ reagent, acetone, 10 min (85% over two steps, 2:1 dr at C3, inseparable mixture of diastereomers); (e) BF3•OEt2, CH2Cl2, 0 °C, 6 h (47% 17+ 29% undesired diketone); (f) 0.4 M KHMDS, PhNTf2, THF, 0 °C, 1 h; (g) 1.2 M Me2Zn, Pd(PPh3)4 (5 mol %), THF, 0 °C to rt, 5 h (84% over two steps). TMSCl, trimethylsilyl chloride; CuTC, copper (I) thiophene-2-carboxylate; PhNTf2, N-phenylbis(trifluoromethanesulfonimide); KHMDS, potassium hexamethyldisilazide; Pd(PPh3)4, tetrakis(triphenylphosphine)palladium.

With the key taxane 6 secured in gram quantities, taxadiene (7) was then targeted, primarily as a means of spectroscopic comparison to previously reported data, but also to demonstrate the feasibility of a large-scale laboratory production of enantioenriched 7. Furthermore, taxadiene (7) is produced in negligible amounts in nature (less than 1 mg can be obtained from 750 kg of tree bark from T. brevifolia)[34] and therefore its optical rotation has never been recorded. To this end, a three-step deoxygenation sequence[48] was performed in 52% yield overall on gram-scale, to generate (+)-7, which was spectroscopically indistinguishable from that of natural 7,[34] previously synthesised (±)-7,[37] and bioengineered (+)-7.[22] The optical rotation of [α]D (CHCl3) = +165° (c = 1.0) is reported herein for the first time and is comparable to that obtained for a bioengineered sample of (+)-7 (see Supplementary Information). This represents the largest quantity of pure (+)-7 isolated to date.

Discussion

In retrospect, the forward synthesis of (+)-taxadienone (6) appears to be rather simple and perhaps intuitive, however, in practice, this optimised approach required many rounds of strategic and tactical revision.[21] Strategically, the retrosynthesis designed in Figure 1C was one of many that were considered at the outset. A small snapshot of the many evaluated blueprints is illustrated in Figure 4. For example, the known difficulties in forming the 6-8-6 tricyclic framework of taxanes[11-20] led us to first consider a ring-closing metathesis strategy to close the central 8-membered ring (disconnection A). However, the realisation that the required substrate 18 would take many steps to build, and the fact that the stereocentres at C1 and C8 would have to be formed with two separate enantioselective reactions, dissuaded us from pursuing this route. An aldol route was then conceived, partly due to the facile formation of 19 via ketone 15 (disconnection B). Although installation of the stereocentres at C1 and C8 might still require two independent enantioselective transformations, the hope was that the reaction conditions used for the aldol cyclisation would concomitantly epimerise the C1 stereocentre into the desired configuration. However, despite a plethora of attempted experiments, the desired cyclisation from 19 did not proceed. Thereafter, strategies involving closure of the AB ring by a Diels–Alder reaction were envisioned. One attempt involved the formation of ketone 20 by a sequence involving a coupling of the enolate of 2,6-dimethylcyclohexanone and a primary alkyl bromide, a Shapiro reaction onto acrolein, oxidation, and Diels–Alder (disconnection C), much akin to the transformation from 14 to 17 (Figure 2). However, ketone 20 already required many steps to construct, and the stereocentre at C8 was challenging to control, despite existing methods in asymmetric enolate alkylation.[49] Lastly, enones 21 were considered as viable intermediates toward the formation of 6 (disconnection D), due to our initial difficulties in forging the C2–C3 bond by Mukaiyama aldol reaction (vide supra). Although enones 21 were formed in short order and already contained all but one carbon of the taxane framework, they did not undergo [4+2] cyclisation, likely due to the rigidity of the sp2 carbons at C3 and C8. Furthermore, a methyl 1,4-addition at C8 was not possible, since reaction first occurred at the less hindered C14, even when a large tert-butyldiphenylsilyl group was appended at C14. After many more strategic revisions and the success in forming the required C2–C3 bond from TMS enol ether 14 using Gd(OTf)3, the final synthetic strategy depicted in Figure 2 was realised.

Figure 4

Initial synthetic investigations toward the synthesis of taxadienone (6)

Disconnection A: An RCM approach would require many more steps in building the taxane framework. Disconnection B: The required aldol closure simply did not proceed. Disconnection C: A Shapiro reaction, followed by aldol and Diels–Alder reactions, is strategically similar to the successful synthetic route, but the stereochemistry at C8 could not be set stereoselectively. Disconnection D: Conjugate addition at C8 to install the methyl unit did not proceed, because only the undesired conjugate addition onto C14 occurred. RCM = Ring-closing metathesis.

Tactically, although quite efficient at present, most steps shown in Figure 2 were initially difficult to scale and suffered from low and inconsistent yields. Even the first transformation from 8 to 10 (Figure 2) was not trivial, despite it being a known two-step transformation.[35,40,41] Modifying the reaction stoichiometry and reaction times was necessary to coax this process into giving good yields consistently on a decagram-scale, with eventual success as a one-pot operation. The second transformation, a merging of the two similar-sized fragments 10 and 11, can be conducted in good yields (86%) on a decagram-scale, but initially, this reaction was plagued with inconsistent yields due to side product formation: the original reaction conditions of tert-butyllithium, BF3•OEt2 and CuI resulted in 1,6-addition of tert-butyllithium (whereas sec-butyllithium is not a very competent nucleophile for this reaction), iodination of 12 (whereas the use of CuBr•SMe2 circumvents this problem), and deconjugation of 12 to give a β,γ-substituted ketone (this problem was rectified by optimising the work-up procedure; see Supplementary Information). The third, asymmetry-inducing step[43] from 12 to 14 was straightforward when run on a small scale (< 100 mg) and delivered high enantioselectivity with 0.5 mol% CuTC and 1 mol% ligand loading. However, upon increasing the reaction scale to gram-scale, the reaction conversion suffered significantly. Eventually, a higher catalyst loading (2 mol% CuTC and 4 mol% ligand) and precise temperature control allowed this reaction to give reliable yields and consistent enantioselectivity on gram-scale. Also worth mentioning is a modified quenching procedure that was developed to address the troublesome TMS trapping of the aluminum enolate[43] (videsupra).

The fourth and the most difficult reaction was the aldol reaction of 14 and acrolein, which only returned desilylated ketone 15 under most reaction conditions. A variety of Lewis acid-mediated reactions (vide supra), as well as anionic silicon-metal exchange reactions, never led to the desired product. This aldol reaction only proceeded when lanthanide triflates such as Yb(OTf)3 or Gd(OTf)3 and very specific solvent systems were used. The final challenge was a scalable Diels–Alder reaction from 16 to 17, which has been known to proceed in moderate yields on similar substrates.[35-37] Although efficient on a small scale, the yield decreased when the reaction was conducted on gram-scale, possibly due to formation of oligomers and polymers. This problem was solved using high dilution conditions (running the reaction at ca. 0.01 M) and through slow addition of substrate 16 to the Lewis acid solution (see Supplementary Information).

Developing a scalable route to the taxane core involved the study and modification of fundamental aspects of the described chemistry, rather than a mere exercise in scale-up. The conciseness of the synthetic route is a direct result of trying to achieve a scalable synthesis, and the reliability of the yields attests to the small variability of reactions ran on a larger scale.[50]

Despite the efficiency of the described approach, there are two obvious limitations in the synthesis of taxadienone (6): 1) the single functional group manipulation from cyclised diketone 17 to the corresponding enol triflate (rendering the route 85% rather than 100% ideal);[42] and 2) the 2:1 diastereoselectivity in the aldol reaction of 14 and acrolein. These issues are currently being addressed by: 1) developing a scalable, one-step enol triflation/methyl coupling reaction; and 2) generating creative acrolein equivalents and/or reaction conditions with various additives to increase to diastereoselectivity of the aldol step.

In summary, a scalable, enantioselective entry to the taxane family of natural products was achieved in only seven steps from commercially available starting material (18–20% overall yield). This triply convergent approach to taxa-4(5),11(12)-dien-2-one (6) allowed for a minimisation of concession steps, wherein six out of seven steps formed skeletal (C–C) bonds. Every one of these steps was performed on a gram-scale, attesting to the scalability and robustness of the sequence. Furthermore, (+)-taxadiene (7) was synthesised, enabling further structural confirmation, as well as the first optical rotation determination of this natural product. The simple chemical route to (+)-7 nicely complements the recent pioneering studies of Stephanopoulos and co-workers,[22] whose bioengineering strategy also delivers gram-scale quantities of (+)-7, albeit as a 9:1 mixture of olefin isomers (taxa-4(5),11(12)-diene (7) and taxa-4(20),11(12)-diene) (see Supplementary Information). Studies are currently underway to utilise the existing functional group handles in 6 to oxidise various sites on the taxane skeleton, in order to create a pyramid-like library of unnatural and natural taxanes en route to Taxol® (1).