The total synthesis of natural products remains an active area of research in chemical synthesis1,2,3. In cases where many congeners of a family of natural products are targeted for synthesis, it is often more efficient to prepare a late-stage intermediate that can be diversified to access the entire collection4,5. In some instances, such late-stage diversification approaches have closely mimicked the biosynthetic pathway to the targeted molecules. For example, congeners of terpenoid secondary metabolites often arise from oxidation or oxygenation reactions that are effected by tailoring P450 enzymes in what has come to be referred to as the oxidase phase6,7. This general approach has been adopted to great effect in preparing many terpenoids8,9,10,11. In our laboratory, we have applied the late-stage diversification approach to the syntheses of members of the longiborneols12,13, the phomactins14,15, the diterpenoid alkaloids16,17, and more recently, cephalotane natural products such as the cephanolides and ceforalides (e.g., 1 and 2) that were prepared from pentacycle 3 (Fig. 1A)18,19.

Fig. 1: Cephalotane natural products.
figure 1

A Benzenoid subfamily and our previous work. B Troponoids and non-aromatic seven-membered ring congeners. C Biosynthesis of the cephalotanes and our strategy employing a single-atom insertion. [O]: oxidation.

The cephanolides20 and ceforalides21 are structurally related to harringtonolide (4), first isolated in 1978 from C. harringtonia (Fig. 1B)22. This natural product has been shown to possess interesting bioactivity, including antiviral and antineoplastic activity23,24. The key difference between these structures is that the cephanolides and ceforalides bear an arene A-ring or oxidized variant thereof (hence our reference to these compounds as the benzenoid congeners), whereas 4 possesses a tropone A-ring. Over the last half-decade, a large number of additional troponoids and non-aromatic seven-membered A-ring cephalotane congeners have been isolated25, including the fortalpinoids (e.g., 5)26, mannolides (e.g., 6)27, and cephinoids (e.g., 7)28. While syntheses of these latter classes of cephalotanes are beginning to appear29,30,31, harringtonolide remains a popular synthetic target32,33,34. Biosynthetically, it is proposed that the benzenoids might be derived from the troponoid subfamily (e.g., 8, Fig. 1C) through a 6π electrocyclization to arrive at the corresponding cyclopropanone (9), which, following a Baeyer–Villiger type oxidation and aromatization, would give the benzenoid type I framework (10). Subsequent decarboxylation and oxidation events would then yield a variety of other congeners bearing the benzenoid type II and III frameworks (11 and 12)20,35. This proposed cephalotane biosynthesis, which relies on a net one-carbon deletion inspired us to explore a contra-biosynthetic approach employing single-atom insertion to prepare the troponoids from the benzenoid subfamily36.

Strategies to achieve such single-atom skeletal edits to access privileged scaffolds continue to emerge and draw the interest of the synthesis community37. Because nitrogen-containing heteroaromatics are the most commonly occurring structural motifs in pharmaceuticals and agrochemicals38,39, many current methods for skeletal editing have relied on the intrinsic reactivity of aza-heterocycles. In our planned approach, we saw an opportunity to highlight skeletal editing of carbocyclic arenes through ring expansion (and ultimately, also ring contraction) to access families of natural products. The challenge of effecting a skeletal change in complex sp3-rich polycyclic structures with multiple functionalities offered opportunities to develop new methods. Here, we report the realization of the benzenoid-to-troponoid conversion of the cephalotanes, culminating in a two-step synthesis of harringtonolide from cephanolide A. Notably, the success of our studies was guided by valuable insights gained through computational analysis of the key ring expansion reaction.

Results and discussion

Harringtonolide (4) has been synthesized by the groups of Mander32, Tang33, and Zhai34 using highly innovative approaches. In particular, Mander’s approach32, which relied on an intramolecular Büchner reaction40,41, was highly inspirational to our planned benzenoid-to-troponoid conversion for the synthesis of 4 (Fig. 1C, Approach A). However, this approach was uniformly unsuccessful even following an extensive survey of reaction conditions (Supplementary Table 1)42,43,44,45,46,47,48. Given the limitations of our attempted intermolecular cycloadditions, we decided to investigate different ring expansion approaches that relied on the reactivity of carbonyl groups (Fig. 1C, Approach B). We were drawn to oxidative dearomatizations of phenols to provide quinols49, which could be followed by a Büchner–Curtius–Schlotterbeck (BCS) reaction50,51,52 to afford the desired tropone moiety (Fig. 2A)53,54,55,56,57,58,59. In general, the BCS reaction has been well-explored and established using saturated ketones60. To the best of our knowledge, there were no reports employing p-quinol derivatives such as 16 as substrates when we carried out these studies. However, during the review of our work, related studies appeared61,62,63. We postulated that a BCS reaction using 16 would undergo ring expansion to 18 via 17. An elimination of the alkoxy group in 18 would yield the desired tropone (19).

Fig. 2: Reaction design for tropone synthesis.
figure 2

A Büchner–Curtius–Schlotterbeck reaction and our hypothesis. B Computational study performed at the ωB97X-D/def2-TZVPP(SMD=CH2Cl2)//ωB97X-D/def2-SVP(SMD=CH2Cl2) level of theory to evaluate the feasibility of tropone formation. LA: Lewis acid, p-: para-.

Reaction design and experimental investigation

To evaluate the feasibility of this tropone synthesis, we have undertaken computational studies as outlined in Fig. 2B64. In the context of the synthesis of harringtonolide (4), we envisioned using quinol methylether 20, which would undergo a one-carbon insertion by a BCS ring-expansion via 21, followed by tautomerization of 22 and loss of methanol (see 22’) to afford 4. We first began our calculations at the several levels of theory (Supplementary Fig. 6A) (Quantum chemical calculations were performed with Gaussian 16 rev. C.01 for geometry optimizations and ORCA 5.0.4 for single-point energy corrections; see the Supplementary Information for full computational details and references.)65,66,67,68 by modeling the reaction of 20 with CH2N2 in the presence of BF3•OEt2, which represents one of the most commonly employed conditions for these types of reactions50. We theorized that the Lewis acid likely binds to the carbonyl lone-pair of 20 away from the α-Me group, as shown in 20’-A. At this stage, two diastereoselective additions of CH2N2 are possible, leading to adducts 21 or 21’, respectively, in which the convex adduct 21 (via TS1; ∆G = 11.7 kcal/mol) is marginally favored by 0.6 kcal/mol. The formation of the tropone ring by ring expansion of 21 is energetically feasible via TS2-A/B (∆G = 6.5 kcal/mol), leading to two constitutional isomers (22 and 23). We also found the possibility of intramolecular oxygen replacement via TS2-C to give rise to epoxide 24. Overall, the C–C migration (TS2) was calculated to be product-determining, wherein a Curtin–Hammett scenario is one of many possibilities to account for our observations.

Given the promising preliminary computational results, we commenced our investigation of the planned BCS reaction by preparing p-quinol derivative 20 (Fig. 3A). Treatment of cephanolide A (1), which was prepared by a modified 12-step sequence (Supplementary Fig. 1), with Kita oxidative dearomatization conditions69,70 afforded 20 in 55% yield. Based on our preliminary calculations, we initially attempted conditions using CH2N2 in the presence of BF3•OEt2 for the tropone formation (Table in Fig. 3A). Unfortunately, these conditions were ineffective and led primarily to the recovery of the starting material (entry 1). Likely, CH2N2 was not nucleophilic enough to react with the carbonyl group of 20 and decomposed under the conditions. Therefore, we turned to other diazomethane equivalents and first examined TMSCHN2 (2.0 equiv) in the presence of BF3•OEt2 (1.2 equiv). To our delight, conducting the reaction at –78 °C yielded tropone 4 but in only 9% isolated yield along with a 57% yield of 25 (a 1:6.3 ratio; entry 2). We also found that using 3.0 equiv of TMSCHN2 at –60 °C, 20 was fully consumed to give 4 in 19% yield and 25 in 70% yield (a 1:3.7 ratio; entry 3). To increase the selectivity for the formation of 4, we then screened a range of Lewis acids (entries 4–9). As a result, we found that AlCl3 (3.0 equiv) along with 5.0 equiv of TMSCHN2 converted 20 to a 37% yield of 4 and 45% yield of 25 (a 1:1.2 ratio; entry 7). Overall, these conditions proved to be optimal (see Supplementary Tables 2 and 3 for full details). Of note, the conversion of 20 to harringtonolide (4) represents the shortest synthesis of this natural product reported to date (14 steps from commercially available material). The selectivity outcome, unexpected based on our preliminary DFT calculations with CH2N2 (entry 3), as well as the improved ratio obtained using AlCl3 (entry 7), led us to undertake additional calculations to gain more insight into the selectivity of this reaction.

Fig. 3: Experimental and computational investigation of the late-stage ring-expansion.
figure 3

A Optimization table for the synthesis of harringtonolide. B Selectivity for the nucleophilic attack of TMSCHN2 on the two prochiral faces of substrate 20-[B]. C Potential energy surface for the reaction between TMSCHN2 and 20-[LA]; All calculations were performed at the ωB97M-V/def2-TZVPP(SMD=CH2Cl2)//ωB97X-D/def2-SVP(SMD=CH2Cl2) level of theory. PIDA: phenyliodine(III) diacetate, TMS: trimethylsilyl, TS: transition state.

Computational studies

With some experimental results in hand, we performed benchmarking computational studies to rationalize the observed selectivity using a range of computational protocols. Based on our computational benchmarking, we found that ωB97M-V/def2-TZVPP(SMD=CH2Cl2)//ωB97X-D/def2-SVP(SMD=CH2Cl2) level of theory most accurately reproduced the empirically observed selectivity. A revised PES has also been calculated for the reaction of 20 with CH2N2 described in Fig. 2B (Supplementary Fig. 6B). Our calculations showed that the attack of TMSCHN2 should occur on the si-face of 20-[LA] — favored by 1 kcal/mol in the case of the BF3-activated substrate (Fig. 3B). However, two possible orientations of the attacking nucleophile are possible, leading to either 26a or 26b. Rearrangement of 26a would yield 27a or 28a, whereas 26b would lead to 27b or 28b. For the computed scenario with BF3•OEt2 as the Lewis acid at –60 °C (Fig. 3C), TS1a-[B] was found to have a 0.5 kcal/mol higher barrier compared to TS1b-[B]. In this case, we believe that the energy difference between TS1a-[B] and TS1b-[B] accounts for the observed distribution of products, which compares favorably with the empirical observation (i.e., the ratio of 4/25 = 1:3.7, which corresponds to a ~ 0.6 kcal/mol difference). With AlCl3 as a Lewis acid, there is no difference in stability between TS1a-[Al] and TS1b-[Al], consistent with our observed ratio (4/25 = 1:1.2). Overall, these computational results show that in the BCS reaction using TMSCHN2, the addition of TMSCHN2 (TS1) to the Lewis acid-bound p-quinol derivative is the selectivity-dictating step.

To gain deeper insight into the impact of the choice of Lewis acid on the reaction outcome, we conducted a comprehensive analysis of the product-determining TSs for both the BF3•OEt2 and AlCl3-mediated systems (Fig. 4). In the case of the minor pathway via TS1a-[B], we observed a C–C bond distance of 2.18 Å between the nucleophilic carbon of TMSCHN2 and the adjacent carbonyl group in an eclipsed orientation. This unexpected, eclipsed orientation of the incoming substituents along the forming C–C bond can be attributed to favorable dispersive interactions between the highly polarizable TMS group and the carbonyl group, as evidenced by the non-covalent interaction (NCI) isosurfaces71. In addition, in the case of TS1b-[B], which features a similar C–C bond distance of 2.19 Å, we observed a staggered orientation of the substituents, with the TMS group placed in close proximity to the BF3 Lewis acid, which sits in the plane of the carbonyl group. This change in orientation from eclipsed to staggered is driven by the favorable interactions between the partially negatively charged fluoride atoms and the electropositive silicon atom, located within 3.26 Å. As such, the preferential reactivity via TS1b-[B] can be attributed to favorable electrostatic and dispersive interactions between the TMS group and the Lewis acid in the case of BF3•OEt2.

Fig. 4: Comparison in the product-determining TSs for reactions mediated by BF3•OEt2 and AlCl3.
figure 4

Effect of the Lewis acid on the product-determining TS of the reaction demonstrated for BF3•OEt2 and AlCl3. The relatively small BF3•OEt2 (left) lies in the plane of the carbonyl group, leading to TS1b-[B] as the favored TS, whereas the larger Lewis acid AlCl3 (right) rotates out of the plane and therefore both TS are equally present. TS: transition state.

When we conducted a similar analysis on the same two competing TSs (i.e., TS1a-[Al] and TS1b-[Al]), using AlCl3 as the Lewis acid, some significant structural differences emerged. Firstly, the forming C–C bonds between TMSCHN2 and the substrate were found to be 0.1 Å longer, which is consistent with earlier TSs, indicating the lower activation energy barriers in this case compared to using BF3 (as shown in Fig. 3). However, a more significant change was observed in the orientation of the AlCl3 group, which moved out of co-planarity with the carbonyl group due to increased steric demand. This fundamental structural alteration eliminates any favorable interaction between the Lewis acid and the TMS group in TS1b-[Al] and promotes the reorientation of the TMS group to a more favorable eclipsed position, similar to that observed in TS1a-[Al]/[B]. Consequently, this loss of favorable non-covalent interactions destabilizes TS1b-[Al], resulting, overall, in better selectivity toward the desired product 27. Finally, in a preliminary study, we have shown that the tropone formation can be extended to other substrates (Supplementary Fig. 2).

In conclusion, we have shown that an oxidative dearomatization and ring expansion starting from cephanolide A accomplishes a benzenoid-to-troponoid ring expansion to afford harringtonolide. To gain insight into the regioselectivity-determining factors in the ring expansion reaction, we have carried out extensive computational studies. These calculations have unveiled the unique effects of the different Lewis acids in establishing secondary interactions with TMSCHN2 which significantly affect the regioselectivity by changing the relative energies of the different transition structures. The extension of the ring expansion transformation described here to other quinol derivatives are provided in Supplementary Fig. 2. Future studies will focus on the application of the Büchner–Curtius–Schlotterbeck transformation to other natural product classes.


General considerations

Commercial reagents and solvents were purchased from Fisher Scientific, Acros Organics, Alfa Aesar, and/or Sigma Aldrich, and used without additional purification. Diazomethane (CH2N2) was generated using an Aldrich® diazomethane-generator with System 45TM. MeCN and MeOH were sparged with argon and dried by passing through alumina columns using argon in a Glass Contour solvent purification system. DCM was freshly distilled over calcium hydride under a N2 atmosphere before each use. Reaction progress was monitored by thin-layer chromatography (TLC) on Macherey-Nagel TLC plates (60 Å, F254 indicator). TLC plates were visualized by exposure to ultraviolet light (254 nm), and/or stained by submersion in aqueous potassium permanganate solution (KMnO4), p-anisaldehyde, or phosphomolybdic acid stain and heating with a heat gun. Organic solutions were concentrated under reduced pressure on a Heidolph temperature-controlled rotary evaporator equipped with a dry ice/isopropanol condenser.

Oxidative dearomatization

To a solution of cephanolide A (1) (25.0 mg, 83.8 μmol, 1.0 equiv) in MeCN/MeOH (1:1 v/v, 838 µL, 0.1 M) was added phenyliodine(III) diacetate (PIDA; 32.4 mg, 101 μmol, 1.2 equiv) at 0 °C under a N2 atmosphere. After stirring at room temperature for 5 h, the reaction mixture was quenched with sat. aq. NaHCO3 (2 mL), diluted with H2O (3 mL) and extracted with DCM (3 × 5 mL). The combined organic phase was dried over Na2SO4 and concentrated in vacuo. The resulting residue was purified by silica gel flash column chromatography (hexanes/EtOAc = 2:1), yielding methyl-ceforalide H (20) (15.2 mg, 46.3 μmol, 55%) as a colorless solid.


A flame-dried vial with a magnetic stir bar was transferred to a glovebox and charged with AlCl3 (12.2 mg, 91.4 µmol, 3.0 equiv). The vial was sealed with a septa cap and removed from the glove box. The vial was evacuated and backfilled with N2 three times and cooled to –60 °C. Freshly distilled DCM (50 µL) was added, and the suspension was stirred at –60 °C for 5 min. A solution of methyl-ceforalide H (20) (10.0 mg, 30.5 µmol, 1.0 equiv) in freshly distilled DCM (250 µL) was added and stirred at –60 °C for 10 min to give a grayish suspension. TMS-diazomethane (0.2 M, prepared from a 2.0 M solution in hexanes diluted with freshly distilled DCM, 760 µL, 152 µmol, 5.0 equiv) was added over 2 min resulting in a yellowish solution. The mixture was stirred at –60 °C for 3 h and quenched with sat. aq. NaHCO3 (500 µL). The suspension was diluted with H2O (2 mL) and extracted with DCM (3 × 3 mL). The combined organic layers were dried over anhydrous Na2SO4, and the solvent was removed under reduced pressure. The residue was purified by preparative TLC (hexanes/EtOAc = 1:3), yielding harringtonolide (4) (3.5 mg, 11.3 µmol, 37%) as a colorless solid and iso-harringtonolide (25) (4.3 mg, 13.9 µmol, 46%) as a colorless solid.

Computational methods

The range-separated dispersion corrected ωB97X-D density functional65 was used in conjunction with the double-zeta valence polarized def2-SVP basis set67, to optimize the geometry of all stationary points. Additional single points energy correction was carried out with the newer generation meta-augmented range separated density functional ωB97M-V9 that employs the Vydrov and van Voorhis VV10 dispersion correction72, together with the triple-zeta valence polarized def2-TZVPP basis set. The VV10 dispersion corrected family of functionals developed by the Head-Gordon group have been demonstrated to be one of the most robust functionals for assessment of main group thermochemistry and for describing non-covalent interactions (Quantum chemical calculations were performed with Gaussian 16 rev. C.01 for geometry optimizations and ORCA 5.0.4 for single-point energy corrections; see the Supplementary Information for full computational details and references.). All calculations included the integral equation formalism variant of the polarizable continuum model (IEF-PCM), with the SMD solvation model to account for solvation effects (solvent = dichloromethane)68. Conformational sampling was performed manually. Gaussian16 version C.01 was employed for all density functional theory (DFT) geometry optimization calculations, using the default ultrafine pruned (99,590) grid for numerical integration of the exchange-correlation functional and its derivatives (Quantum chemical calculations were performed with Gaussian 16 rev. C.01 for geometry optimizations and ORCA 5.0.4 for single-point energy corrections; see the Supplementary Information for full computational details and references.). Single point corrections were carried our using ORCA 5.0.473. Vibrational frequency calculations were used to verify that stationary points were either minima or first-order saddle points on the corresponding potential energy surface. Additional intrinsic reaction coordinate (IRC) calculations were performed to ensure that the transition state structures connected to their appropriate initial and final geometries74. The computed thermochemistry data were further corrected following Grimme’s quasi-harmonic (QHA)75 model for entropy with a frequency cut-off value of 100.0 cm−1 using the GoodVibes76 program at 213.15 K (−60° C). In addition, GoodVibes applied 1 M standard concentration corrections to all individual calculations to account for reactions in solution (i.e., change in standard concentration from 1 atm to 1 M)77. XYZ coordinate files were also generated using GoodVibes.