Convergent Total Synthesis of (−)-Cyclopamine

Abstract

A concise and enantioselective total synthesis of the Veratrum alkaloid cyclopamine is disclosed. This highly convergent synthesis with a 16-step longest linear sequence (LLS) was enabled by a de novo synthesis of the trans-6,5 hetero-bicycle via a strain-inducing halocyclization process, a key Tsuji-Trost cyclization to construct the fully substituted, spirocyclic THF motif with exquisite diastereocontrol, and a late-stage ring-closing metathesis (RCM) reaction to forge the central tetrasubstituted olefin.

Graphical Abstract

Cyclopamine (1), first isolated from Veratrum californicum (California corn lily),¹ is a notable member of the Veratrum alkaloid family (Figure 1A).² It is known ex post facto for inducing the cyclopia birth defect in sheep, where the affected new-born lambs only had one eye and were reminiscent of the cyclops from Greek mythology.³ Studies from Beachy et al. revealed that cyclopamine is a potent inhibitor of the highly conserved hedgehog signaling pathway, which is critical for the correct differentiation and symmetry development of embryos.⁴ Despite its known decomposition pathway leading to veratramine (3) or potentially other byproducts under acidic conditions,⁵ cyclopamine has been recognized as a promising lead compound for anti-cancer agents and numerous medicinal analogs have been reported to date, most notably the semi-synthetically prepared saridegib (4).⁶

Figure 1.

Cyclopamine (1): (A) related family members, medicinal analogs, and previous syntheses; (B) retrosynthetic analysis in this work.

The intricate structures of Veratrum alkaloids have piqued the interest of synthetic chemists for years.⁷ With respect to cyclopamine, the rare C-nor-D-homo steroid skeleton is further complicated by the fully substituted, spirocyclic THF motif, the unique trans-6,5 EF ring system, and the central tetrasubstituted olefin, all of which posed unique challenges to its synthesis. A pioneering synthesis of the closely related jervine (2) by the Masamune group involves an 18-step sequence starting from the advanced intermediate 6, which in turn must be prepared in 25 steps from Hagemann’s ester or obtained from degradation of hecogenin.^7a More recently in 2009, Giannis et al. disclosed an elegant semi-synthesis of cyclopamine in 20 steps from dehydroepiandrosterone (5), featuring a biomimetic 1,2-skeletal rearrangement to construct the C-nor-D-homo steroid scaffold.^7h Multiple progress-towards studies of jervine or cyclopamine have also been reported to date.⁸ This Communication discloses a convergent and enantioselective total synthesis of 1 with a 16-step longest linear sequence (LLS).

The final retrosynthetic depiction shown in Figure 1B benefitted from the learnings of multiple generations of unsuccessful routes, most of which were based on the convergent union of ABC and EF fragments followed by D-ring formation post-coupling (see SI for details). Learnings from those studies led to the current approach wherein a fully formed ABC tricycle (Fragment A, 7) was coupled to an EF bicycle progenitor (Fragment B, 8) via a diastereoselective 1,2-addition. The late-stage formation of the D/E ring systems hinged on a Tsuji-Trost cyclization to forge the THF ring and an RCM reaction to build the central tetrasubstituted olefin. 7 and 8 could in turn be prepared from readily available building blocks such as (S)-Wieland Miescher ketone (9) and 2-siloxyfuran 10.

The synthesis of 7 (Scheme 1A) commenced from the decalin 11, the TBS variant of which was previously prepared in 4 steps from (S)-Wieland-Miescher ketone (9).^8k Unfortunately, our attempts to directly adapt this procedure for the preparation of Bn-protected 11 were unsuccessful (see SI) and eventually required 2 additional steps to exchange the protecting group, resulting in a non-ideal 6 step sequence. An alternative 3-step protocol amenable to gram scale operations was thus developed involving enone isomerization/reduction of acetal protected 9 (see SI for details). A cyclopentenone annulation sequence was subsequently carried out using the known phosphonate reagent 12⁹: the lithium enolate of 11 was first alkylated with 12 to afford intermediate 13 as a 1:1 mixture of C9 epimers (after hydrolysis of the enol ether), which upon treatment with Cs₂CO₃ underwent an intramolecular Horner-Wadsworth-Emmons reaction to forge the cyclopentenone motif with concomitant equilibration to a 9:1 mixture of C9 epimers favoring the desired diastereomer, as confirmed by X-ray crystallographic analysis. Further enrichment of 14 to >20:1 dr was achieved after a single round of recrystallization from hexanes/DCM (ca. 10:1). With gram quantities of 14 at hand, a second enolate alkylation at C13 with tert-butyl bromoacetate proceeded smoothly to yield 15 as a single diastereomer. The conjugate reduction of enone 15 required judicious choice of reagents (see SI for the full optimization table). Whereas Stryker’s reagent and SmI₂ gave no reaction and Shenvi’s HAT reduction conditions¹⁰ resulted in decomposition, a reactive CuH species reported by Lipshutz and coworkers¹¹ proved to be the optimal choice, furnishing the desired 1,4-reduction product 16 in 86% yield as a single diastereomer without any undesired 1,2-reduction. The excellent diastereoselectivity observed could be rationalized by the angular methyl group blocking CuH approach from the top face of 15. Ketone 16 was ultimately transformed to the 1,2-addition precursor 7 by Wittig olefination, DIBAL-H reduction of the tBu ester, and finally an Appel reaction to convert the resulting primary alcohol to iodide.

Scheme 1. Synthesis of fragments A (7) and B (8).^a.

^a For detailed reagents and conditions, see Supporting Information.

Multiple strategies (see SI) to access fragment B (8) were evaluated via a structure such as lactone 20 (Scheme 1B). Similar molecules have been prepared in the past through labor-intensive chiral pool strategies. For instance, a derivative of 20 bearing a Cbz instead of a benzenesulfonyl (Bs) group was previously accessed in either 9 steps^8c or 6 steps^8h from (S)-citronellal, with 2 additional steps required for protecting group exchange as the Bs group has proven to be essential for late-stage manipulations (see SI for discussion). A concise 4-step route to 20 was thus developed through an unusual sequence relying on the C5-selective asymmetric allylic alkylation (AAA) of siloxyfuran 10 as reported by Arseniyadis et al.¹² This key precedent was improved for the synthesis of γ-butenolide 17 through the addition of NH₄OAc (as per suggested by the Arseniyadis group in our correspondence), which further boosted reaction efficiency and regioselectivity (favoring C5 allylation over C3), delivering 17 in 75% isolated yield and 97% ee (see SI for details). Subsequent aza-Michael addition to 17 required substantial screening (see SI) that eventually led to the identification of BsNHOBn (N-benzenesulfonyl-O-benzylhydroxylamine) as the optimal nucleophile. X-ray analysis confirmed that addition occurred exclusively trans to the existing γ substituent on the butenolide. In its fully optimized form, this reaction involved directly treating the aza-Michael adduct with Zn⁰/NH₄Cl in the same reaction vessel to effect N-O bond cleavage and furnish the sulfonamide 18 as a single diastereomer in 71% isolated yield. Racemization of 18 (71% ee) was observed but could be rectified by a single round of recrystallization from DCM/Et₂O (1:1), which restored its enantiopurity (>97% ee, 50% yield overall). With gram quantities of 18 at hand, the key halocyclization step to construct the strained trans-6,5 EF bicycle was investigated. Due to the extreme strain that would be incurred through a 5-exo-trig cyclization to deliver a trans-5,5 ring system, it was rationalized that an unusual 6-endo-trig cyclization would be preferred. After the initial hit with NIS/K₂CO₃ verified this hypothesis, further screening was performed (see SI for details) that eventually led to the most reproducible and scalable protocol with KOtBu/I₂. In its fully optimized form, the halocyclization step furnished the tertiary iodide 19 in 51% yield as an inconsequential 10:1 mixture of diastereomers (favoring the axial iodide, see SI), which was then subjected to a tin hydride-mediated dehalogenation reaction to afford the desired lactone 20 in 71% yield as a single diastereomer. This stereochemical outcome could be rationalized as a combination of stereoelectronics (axial iodide is more susceptible to abstraction), thermodynamic preference (methyl group prefers to be equatorial), and steric bulk of the benzenesulfonyl group (which disfavors hydride approach from top face of the tertiary radical). It is worth noting that the ring-opening side product 18 was consistently observed on larger scale (>100 mg) and could be isolated in 21% yield. Attempts to suppress this process by running the reaction at ambient temperature instead of 100 °C (see SI) surprisingly exacerbated ring-opening (20:18 = 1:1, vs. 3:1 with AIBN at 100 °C). Besides this piece of evidence, we also observed that 1) the halocyclization reaction exhibited certain degrees of reversibility over prolonged reaction times (see SI), and 2) addition of TEMPO to the halocyclization reaction inhibited product formation. These clues led us to speculate that the halocylization might proceed through a one-electron pathway involving formation and homolysis of an N-I species (see SI). Methylation of 20 by enolate alkylation proceeded smoothly to deliver the key lactone 21 as a single diastereomer, the structure of which was confirmed by X-ray analysis. The diastereoselectivity observed was in line with Taber’s report^8h on a similar scaffold and could be attributed to the axial β-H at the ring junction blocking electrophile approach from the top face of 20. Finally, addition of isopropenyllithium to lactone 21 yielded a hemiketal that was in equilibrium with the open-chain hydroxyketone, the secondary alcohol of which could be TES protected in the same pot to furnish enone 8 in 65% yield.

With both fragments (7 and 8) in hand, the endgame sequence (Scheme 2) commenced with union of the two fragments by 1,2-addition, wherein the primary iodide 7 first underwent lithium halogen exchange with tBuLi, and then the resulting alkyllithium species added into enone 8. Addition of LaCl₃•2LiCl proved to be crucial (see SI for additive screening) for the exclusive 1,2-regioselectivity observed in this reaction,¹³ yielding tertiary alcohol 22 as a single diastereomer (confirmed by X-ray) possibly due to Felkin-Anh selectivity. This success set the stage for the final D/E ring double cyclization sequence, which entailed formation of the E ring through an alkoxide Tsuji-Trost cyclization followed by formation of the D ring by ring-closing metathesis (RCM). The challenging Tsuji-Trost cyclization turned out to be an immense undertaking that required considerable optimization (see SI). To our delight, Boc carbonate 23 (prepared from 22 in two steps) emerged as the singularly successful substrate, delivering the desired THF-containing RCM precursor 24 in 73% yield with perfect net stereoretention from tertiary alcohol 22. This diastereochemical outcome was in line with previous observations in similar THF-forming Tsuji-Trost cyclizations.¹⁴ The RCM reaction to form the tetrasubstituted olefin, which was the final hurdle in this synthesis, was initially met with expected difficulties (see SI for details): conventional metathesis catalysts such as Hoveyda-Grubbs 2^nd generation catalyst (Ru-1) gave low conversions even at elevated temperatures, and the less sterically encumbering o-tolyl variant Ru-2 only led to slightly improved results. Fortunately, the indenylidene catalyst Ru-3 was reported to be an extremely effective metathesis catalyst for sterically demanding substrates.¹⁵ Indeed, a significant improvement in yield was observed with Ru-3, which, when combined with perfluorotoluene as solvent, ^15b led to the optimal condition that afforded bis-protected cyclopamine 25 in 85% yield, the structure of which was confirmed by X-ray analysis. The prior approach of Giannis necessitated two distinct reductive deprotection conditions to excise Bn and sulfonamide groups in 79% overall yield.^7h Instead, removal of both groups could be accomplished with freshly prepared LiDBB¹⁶, yielding (−)-cyclopamine in 78% yield. It is worth noting that when acquiring NMR data for cyclopamine in CD₂CL₂, a solvent that was used in multiple previous reports, we observed that concentration of the sample had a significant impact on the chemical shifts of cyclopamine’s ¹H and ¹³C spectra (see SI for discussion).

Scheme 2. Total synthesis of (−)-cyclopamine (1).^a.

^a For detailed reagents and conditions, see Supporting Information.

To conclude, a convergent total synthesis of (−)-cyclopamine (1) was achieved with a 16-step LLS (1.4% overall yield, 61% ideality,¹⁷ see SI for calculation) from (S)-Wieland-Miescher ketone. Unlike the semi-synthetic routes which predate this disclosure, the strategy outlined herein should be amenable to exploring deep-seated structural modifications for further SAR studies of this promising class of natural products. Key areas for improvement for which there are methodological gaps include a more direct conversion of 18 to 20 (see SI for discussion) and a more concise fragment coupling that can minimize protecting group manipulations. That said, notable features of this synthesis include rapid stereocontrolled access to similarly sized coupling fragments, namely the streamlined ABC-ring synthesis through annulation/CuH reduction and an unusually concise furan-based access to the key EF ring system via AAA/aza-Michael/halocyclization. Furthermore, the exquisite stereocontrol exhibited in the fragment coupling/strained THF ring formation prior to the high-yielding RCM cyclization represents a bold yet effective means of accessing 1.