Introduction

In 1986, Uemura reported the isolation and structural elucidation of the polyether macrolide halichondrin B (1, Fig. 1A)1 from the sponge Halichondria okadai. This structurally complex analog of the known macrolide norhalichondrin A2 demonstrated ‘extraordinary’3 activity (IC50 = 93 pg/mL against B-l6 melanoma cells) and in mouse models of P-388 leukemia increased median survival time by more than threefold at doses as low as 10 μg/kg1. To support further biological testing, Kishi and co-workers reported a landmark total synthesis of halichondrin B in 1992, a remarkable effort that required more than 100 synthetic transformations3. Synthesis of (nor)halichondrin B reported by Phillips4 and Nicoloau5,6, and synthetic studies toward the halichondrins reported by Salomon7, Horita and Yonemitsu8, Burke9, and others4 have also provided valuable insight and strategies for assembling these complex polyketides. Importantly, the biological testing of synthetic intermediates in the Kishi process provided critical insight into the relationship between structure and anticancer activity within the halichondrins10. Most notably, it was found that the activity of halichondrin B relied only on the right-hand portion of the molecule (C1–C36). Based on these findings, subsequent and extensive synthetic work at Eisai led to the development of eribulin (Halaven: 2), which represents a simplified version of halichondrin B whereby the C35–C54 fragment of the natural product was removed and the lactone oxygen replaced by a methylene group (see the boxed region of 2)11. Biological characterization of eribulin has since revealed a unique mechanism of action that involves binding to the growing end of microtubules, disruption of microtubule dynamics, and ultimately irreversible mitotic arrest and cell death by apoptosis4,12. Following successful clinical trials, eribulin was approved by the US Food and Drug Administration for use in pretreated metastatic breast cancer in 2010 and subsequently for the treatment of inoperable liposarcoma in 2016.

Fig. 1: Eribulin: origins, Eisai and Alphora process synthesis intermediates and α-chloroaldehyde building blocks for eribulin.
figure 1

A The anticancer marine natural product halichondrin B and synthetic analog eribulin. B Key building blocks used in the Eisai and Alphora process synthesis of eribulin. C Key disconnections in our proposed eribulin synthesis rely on α-chloroaldehydes generated via organocatalysis as the predominant source of chirality for the C14–C35 fragment. Red arrows depict an epoxide opening/rearrangement reaction. The green bar denotes bonds formed by chloride displacement. Blue X represents O or NBoc. PG protecting group, Piv pivalate, Boc tert-butoxycarbonyl, TBS tert-butyldimethylsilyl.

Full size image

Structurally, eribulin represents the most complex non-peptidic and fully synthetic drug, with 19 stereocenters, 3 tetrahydrofuran and 3 tetrahydropyran rings, and a 22-membered macrocyclic ketone12. Owing to the myriad challenges in synthesizing a molecule of this complexity, extensive efforts have focused on improving eribulin synthesis. Much of this work is predicated by several generations of halichondrin syntheses reported by Kishi3,13,14,15. In 2013, the Process Research Group at Eisai disclosed their multi-gram synthesis12 and commercial manufacturing route16,17,18 for eribulin that involved a late-stage union of the sulfone function in 3 and aldehyde in 7 (Fig. 1B) followed by a Nozaki-Hiyama-Kishi (NHK)19,20 coupling to close the macrocyclic ring. Completion of the synthesis then involved the formation of the C14 ketal function and conversion of the C35 alcohol into the requisite amine. The commercial manufacturing route requires a total of 67 steps, with 33 steps in the longest linear sequence. The majority (43) of these steps are required to assemble the C14–C35 fragment 3, which contains 10 chiral centers and three densely functionalized tetrahydrofuran/pyrans. Notably, while the four chiral centers found in the C29–C32 tetrahydrofuran derive from D-glucurono-3,6-lactone, controlling both the relative and absolute stereochemistry at the remaining six chiral centers in 3 requires one chiral separation using simulated moving bed chromatography, as well as four separate asymmetric reactions, two with stoichiometric chiral ligand16. An alternative process for eribulin production has been reported by the Process Research Group at Alphora and involves the early introduction of the C35 amine group and the advanced intermediate 4, which is converted into eribulin by a process similar to that used in the Eisai synthesis21,22. Notably, this eribulin synthesis requires a total of 73 steps, with 34 steps in the longest linear sequence of reactions. In both the Eisai and Alphora routes, assembly of the C14–C35 fragment involves an NHK coupling between the key intermediates 5 and 6, followed by a cyclic etherification reaction to construct the C23–C27 tetrahydropyran. Kishi has also reported several alternative strategies for accessing key eribulin building blocks and formation of the macrocyclic ketone15,23,24,25,26, including a convergent Pd(0)-mediated macroketocyclization strategy15. When applied to eribulin, this later approach reduces the number of linear steps but requires a similar number of total steps to that of the Eisai process. Very recently, the Nicolaou group reported a synthesis of eribulin that exploits a similar NHK coupling-pyran formation strategy but involves the prior formation of a fully intact C1–C26 fragment27 and builds on their efforts directed toward the halichondrins5,6. In this work, the macrocyclic ring is constructed through a Co/Cr catalyzed intramolecular coupling between an alkyl iodide and an aldehyde function at C1. Our longstanding interest in the use of α-chloroaldehydes28,29,30,31,32 in tetrahydrofuran33,34,35 and natural product synthesis36,37,38 suggested that these highly versatile and readily prepared building blocks would be particularly well-suited for an eribulin synthesis. Here, we show convergent syntheses of the C14–C35 fragment of eribulin used in both the Eisai and Alphora synthetic routes using α-chloroaldehyde building blocks. These syntheses also exploit a regio- and diastereoselective Corey–Chaykovsky reaction to connect the two tetrahydrofuran subunits. This work highlights the use of α-chloroaldehydes in complex molecule synthesis and significantly reduces the total number of reactions required to access this important cancer therapeutic.

Results

An α-chloroaldehyde-based plan for synthesizing eribulin

As outlined in general terms in Fig. 1C, our α-chloroaldehyde-based strategy to access eribulin would rely on an elaborate Corey–Chaykovsky reaction39 to affect the union of the C14–C26 ketone 10 and C27–C35 sulfonium salt 9 (X = O or NBoc). This key coupling reaction would necessarily involve regioselective deprotonation of either of the sulfoniums 9 at C27 and a subsequent doubly diastereoselective addition to ketone 10, thus establishing the C26 and C27 stereocenters. Regioselective rearrangement of the resulting C26–C27 epoxide would then set the stage for the formation of the C23–C27 tetrahydropyran via chloride displacement23 and ultimately converge with the Eisai or Alphora intermediates 3 or 422. The two tetrahydrofuran units in 4 should themselves be accessible from readily available α-chloroaldehydes 11 and 13. Likewise, the C23 chloride function in 10 that is eventually required for tetrahydropyran formation would also be derived from an α-chloroaldehyde (e.g., 12) via a sequence involving a Horner–Wadsworth–Emmons (HWE) reaction40 and subsequent enone reduction. Following this strategy, each of the nine stereocenters found in the C14–C33 fragment of eribulin would be introduced using substrate-based stereocontrol starting from one of the α-chloroaldehydes 11–13.

Synthesis of the C27–C35 sulfonium

Toward the synthetic goals outlined above, we first examined the proline-catalyzed coupling of tetrahydro-4H-thiopyran-4-one with the aldehyde 16, which is available in five steps from commercially available (S)-tetrahydrofufurylamine (14) (Fig. 2). This one-pot process involves several key reactions all orchestrated by proline35. First, an α-chlorination of the aldehyde produces a mixture of α-chloroaldehydes (2R)-17 and (2S)-17. Importantly, while the α-chlorination is not stereoselective, proline also promotes epimerization of these diastereomeric α-chloroaldehydes and the subsequent proline-catalyzed aldol reaction with tetrahydrothiopyranone41 is sufficiently slow to effect a dynamic kinetic resolution, favoring the reaction of (2S)-17 and formation of the anti-aldol syn-chlorohydrin 19. Here, using our optimized reaction conditions41, the chlorohydrin 19 could be readily prepared on multi-gram scale as a single diastereomer, albeit in modest yield (35–40%) along with ~40% of a 1:1 mixture of α-chloroaldehydes (R)- and (S)-17 (67% yield of 19 based on recovered 17). Careful monitoring of this reaction indicated that off-cycle hemiaminals42 derived from the reaction of proline with the intermediate α-chloroaldehydes 17 prevented further conversion. To improve on this outcome, we carried out an exhaustive screen of solvents and additives and eventually found that either five equivalents of dimethylsulfone (DMSO2) or 0.6 equivalents of PPTS43 facilitated hydrolysis of the off-cycle hemiaminals. Using these conditions, the conversion improved significantly (59%); however, the diastereocontrol decreased, and the aldol product 19 (41% yield) was accompanied by the anti-chlorohydrin 18 (18% yield). Even accounting for the formation of 18, these conditions proved favorable for advancing larger amounts of material. A subsequent diastereoselective reduction using DIBAL followed by a SrCO3 promoted cyclization15, then gave the tetrahydrofuran 20. From here, Mitsunobu inversion44 followed by hydrolysis gave the correctly configured tetrahydrofuranol 21. Finally, methylation of the free alcohol and arylation of the thioether function using a diaryliodonium salt provided the key sulfonium Corey–Chaykovsky coupling partner 22.

Fig. 2: Synthesis of the C27–C35 sulfonium 22.
figure 2

A one-pot proline-catalyzed α-chlorination/aldol reaction provides rapid access to the chlorohydrin 19, which can be converted into the sulfonium salt 22 in preparation for the key Corey–Chaykovsky coupling. acac acetylacetonate, Pro proline, NCS N-chlorosuccinimide, DIBAL diisobutylaluminium hydride, PPTS pyridinium para-toluenesulfonic acid.

Full size image

Synthesis of the C14–C26 ketone

Synthesis of the ketone 33 started with an asymmetric α-chlorination of the readily available aldehyde 23 using Christmann’s modification31 of the MacMillan α-chlorination reaction (Fig. 3)29. Here, we found that the addition of pentane to the crude reaction mixture (in acetonitrile) allowed for direct extraction of pure (>95%) α-chloroaldehyde 25, which was produced in 95% ee. With the α-chloroaldehyde 25 in hand, a lithium aldol reaction33 with the enolate derived from methyl ketone 26 (made in one step from commercially available 5-hydroxy-2-pentanone) gave the β-hydroxy ketochlorohydrin 27 (d.r. = 4:1). The ketone function in this material was then reduced in a 1,3-syn-selective manner using DIBAL to afford the corresponding diol (not shown). At this point, we examined both the thermal (MeOH, 120 °C, μwave)34 and silver(I)-promoted (AgOTf, Ag2O, THF)33 cyclization conditions reported by us as well as the SrCO3-promoted cyclization protocol more recently reported by Kishi15. In this case, the Ag(I)-promoted cyclization conditions proved superior and delivered the desired tetrahydrofuranol that was protected as the corresponding TBS ether 28 in excellent overall yield. Oxidative cleavage of the alkene function followed by a second α-chlorination29 using the MacMillan catalyst ent-24 gave the α-chloroaldehyde 30. Notably, this reaction also upgrades the enantiomeric purity to >99% ee.

Fig. 3: Synthesis of the C14–C26 methyl ketone 33.
figure 3

Two separate asymmetric α-chlorination reactions are employed in the synthesis of the chloroketone 33 required for the key Corey–Chaykovsky coupling. LDA lithium diisopropyl amide, dpm dipivaloylmethanato, DBU 1,8-diazabicyclo[5.4.0]undec-7-ene.

Full size image

To complete the sequence outlined in Fig. 3, an HWE reaction involving the α-chloroaldehyde 30 gave the enone 31. Unfortunately, reduction of the alkene function in 31 proved to be problematic and largely delivered dechlorinated products using established protocols for the conjugate reduction of enones (e.g., catalytic hydrogenation45, Ni-mediated hydride transfer46, or CuH reduction47). To avoid reduction of the chloromethine function, we prepared several model γ-chloroenones and explored a wide range of reduction conditions. From these studies, only the Mn(III)-catalyzed conjugate reduction using phenylsilane and reported by Magnus48 delivered saturated γ-chloroketones in good yield. Reduction of enone 31 using these conditions gave the C25 epimeric ketones 32 and 33 as a 1:1.5 mixture. Although we were unable to improve upon this ratio, the undesired diastereomer could be readily separated by flash column chromatography and treated with DBU in MeCN, which gave a clean 1:1 mixture of diastereomers 32 and 33. Thus, after two recycling events, we could routinely produce significant quantities of the ketone 33.

Synthesis of the C14–C35 Alphora intermediate

With our two building blocks 22 and 33 in hand, we were ready to explore the key Corey–Chaykovsky39 coupling (Fig. 4). Pleasingly, discrimination between the two methylenes adjacent to the sulfonium function in 22 at C27 and C30 was possible using LiHMDS, which effected a regioselective deprotonation of the more sterically accessible protons at C27. The reaction of the resulting ylide with the ketone function in 33, followed by direct oxidation to the corresponding sulfone afforded the epoxide 34 as the major product in good yield. Based on the observed diastereoselectivity in the coupling reaction, we propose that addition49 of the sulfonium ylide from the bottom (Re) face (as shown for 36, top inset, Fig. 4) ultimately controls the stereochemistry at C27. The addition to the ketone function is then governed by Felkin control50, with the C25 methyl-bearing stereocenter dictating the stereochemical outcome at C26. Delighted by this result, we examined several regioselective deprotonation/epoxide opening processes on a series of related model substrates and found the ideal conditions involved the use of a titanocene (III) complex (Cp2TiCl)51, which promotes the formation of an intermediate β-titanoxy tertiary radical52. Reduction of the tertiary radical by a second equivalent of Ti(III) then affords a titanium carbanion that can undergo β-hydride elimination to afford, after protonation, the rearranged allylic alcohol. Applying these conditions to the more elaborate epoxide 34, we were pleased to find that the allylic alcohol 35 was produced as the major product in excellent yield. From here, cyclization to the tetrahydropyran was accomplished using double displacement conditions reported by Kishi on a related precursor to eribulin23. Here, displacement of the chloride occurs through initial attack by the C17–C20 tetrahydrofuran ring oxygen to form a bicyclic oxonium ion 37 that is subsequently attacked by the C27 alcohol function to form the correctly configured tetrahydropyran. Completion of the synthesis of the C14–C35 sulfone 4 then involved removal of the silyl protecting group and oxidation to the ketone 38 followed by olefination and removal of the pivaloyl (Piv) protecting group from the C14 alcohol. The spectroscopic data recorded on our sulfone 39 was identical to that reported previously by researchers at Alphora for the equivalent material (see inset)22. Remarkably, this α-chloroaldehyde-based approach to the key eribulin fragment 4 requires only 30 steps in total, which is a significant (~40%) reduction from that required in the route reported. Conversion of the sulfone 4 into eribulin has been reported by researchers from Alphora and involves an eight-step sequence of reactions. Thus, this work constitutes a formal synthesis of eribulin.

Fig. 4: Synthesis of the C14–C35 Alphora sulfone 4.
figure 4

A doubly diastereoselective Corey–Chaykovsky reaction connects the sulfonium salt 22 and chloroketone 33. A subsequent regioselective epoxide opening sets the stage for the formation of the C23–C27 tetrahydropyran ring. The inset depicts an overlay of 1H NMR spectra (3.0–5.0 ppm only) recorded on the sulfone 39 and the equivalent material produced via GMP process (spectra kindly provided by Alphora). Blue and red text describe facial selectivity in the Corey–Chaykovsky reaction. LiHMDS Lithium bis(trimethylsilyl)amide, MCPBA meta-chloroperoxybenzoic acid, DMP Dess–Martin periodinane, DTBMP 2,6-di-tert-butylpyridine.

Full size image

Synthesis of the C14–C35 Eisai intermediate

Having demonstrated the synthesis of protected amino alcohols 4 and 39, we further exploited this process for the production of the protected diols 3 and 46 that represent intermediates in both the Eisai commercial route53 for eribulin production and synthesis of eribulin reported by Kishi, respectively23. As summarized in Fig. 5, commercially available furanone 40 was converted into aldehyde 41, which was then subjected to the same series of reactions described in Fig. 2 for aldehyde 16 (see Supplementary Information for details) to provide the arylsulfonium 42. A subsequent Corey–Chaykovsky reaction with ketone 33 followed by oxidation of the thioether gave sulfone 43 in good yield. Conversion of this epoxide into the desired pyran 45 proceeded smoothly using the conditions developed for the protected amino alcohol 35 (Fig. 4). Finally, olefination gave the acetonide-protected C14–C35 fragment 46 that has been previously reported by Kishi as an intermediate in an eribulin synthesis. Deprotection of the diol function followed by reaction with TBSOTf and base then gave the bis-TBS-protected diol 3 used in the reported commercial route for eribulin production by Eisai. Notably, these routes require only 27–29 steps in total, which represents ~30% reduction in steps required to access this key eribulin intermediate. The spectral data derived from both 46 and 3 agreed with that reported for these compounds.

Fig. 5: Synthesis of Kishi and Eisai intermediates 46 and 3.
figure 5

Exploiting the Corey–Chaykovsky coupling and Ti(III)-promoted rearrangement sequence, synthesis of the known acetonide- and TBS-protected diols 46 and 3 was accomplished. Cp cyclopentadienyl.

Full size image

Discussion

A synthetic strategy has been developed that exploits the inherent stereochemistry of three readily prepared α-chloroaldehydes to control the relative and absolute stereochemistry at 9 of the 10 stereogenic centers in the C14–C35 fragment of eribulin. In addition to orchestrating a series of diastereoselective reactions, each chlorine atom is ultimately displaced in the formation of one of the three heterocyclic rings and thus serves two key roles. A doubly diastereoselective Corey–Chaykovsky reaction was also critical for both securing the key C27 stereocenter and uniting our two α-chloroaldehyde-derived building blocks: sulfonium salts 22 and 42 and ketone 33. Overall, these efforts constitute formal syntheses of eribulin that merge with reported routes to this important anticancer agent and reduce the total number of reactions required to produce eribulin to 52, which compares well with that required for either the Eisai (67 steps)16 or Alphora (73 steps)22 processes, and reduces the longest linear sequence to 28 steps. While it is difficult to compare processes run on kilogram16 to gram22 to milligram (this work) scale, several aspects of the process presented here may prove advantageous, including a decreased reliance on chiral catalysts and chiral chromatography, low temperature (<0 °C) reactions (5 vs. 10), and the obvious benefits derived from overall improvements to step economy54. Perhaps of equal importance, the present syntheses showcase the unique versatility of α-chloroaldehydes as chiral building blocks for constructing the types of oxygen-containing heterocycles that are often encountered in structurally complex polyketide natural products. We expect these efforts will inspire further use of α-chloroaldehydes in total synthesis and may well support drug discovery and development efforts.