Structure Revision of the Lomaiviticins

Abstract

The lomaiviticins are dimeric genotoxic metabolites that contain unusual diazocyclopentadiene functional groups and 2–4 deoxyglycoside residues. Because only 6 of 19 carbon atoms in the monomeric aglycon unit are proton-attached, their structure determination by NMR spectroscopic analysis is difficult. Prior structure elucidation efforts established that the two halves of the lomaiviticins are joined by a single carbon–carbon bond appended to an oxidized cyclohexenone ring. This ring was believed to comprise a 4,5-dihydroxycyclohex-2-ene-1-one. The bridging bond was positioned at C6. This structure proposal has not been tested because no lomaiviticin has been prepared by total chemical synthesis or successfully analyzed by X-ray crystallography. Here, we disclose microED studies which establish that (−)-lomaiviticin C contains a 4,6-dihydroxycyclohex-2-ene-1-one residue, that the bridging carbon–carbon bond is located at C5, and that the orientation of the cyclohexenone ring and configuration of the secondary glycoside are reversed, relative to their original assignment. High-field (800 MHz) NMR analysis supports the revised assignment and suggests earlier efforts were misled by a combination of a near-zero ³J_H4,H5 coupling constant and a ⁴J_C,H coupling interpreted as a ³J_C,H coupling. DFT calculations of the expected ¹³C chemical shifts and C–H coupling constants provide further robust support for the structure revision. Because the interconversion of lomaiviticins A, B, and C has been demonstrated, these findings apply to each isolate. These studies clarify the structures of this family of metabolites and underscore the power of microED analysis in natural product structure determination.

Graphical Abstract

INTRODUCTION

The lomaiviticins are cytotoxic bacterial metabolites characterized by the presence of the deoxyglycosides β-N,N-dimethyl-l-pyrrolosamine and α-l-oleandrose, one or two diazotetrahydro[b]benzofluorene (diazofluorene) residues, and a single carbon–carbon bond linking two highly oxygenated cyclohexene rings (Figure 1).¹ The C₂-symmetric derivatives known as (−)-lomaiviticins A and B were first reported in 2001. The structures 1a and 2a were advanced by He and co-workers based on HRMS and extensive NMR analysis (Figure 1a).² The relative stereochemistry (including anomeric configuration) of the carbohydrates was determined by analysis of ³J_H,H coupling constants and by comparison to other natural glycosides.³ The presence of the diazofluorene was inferred from the kinamycins (e.g, (−)-kinamycin F (4)), monomeric diazofluorenes whose structures are known with certainty.⁴ The structure elucidation of the 2,2′-bicyclohexyl core was more challenging. In aggregate, only 6 of 19 carbon atoms in the monomeric aglycon unit are proton-attached, and two of these carbon atoms (C8, C9) are separated from the core by the diazofluorene. Thus, C2, C4, C12, and C13 constitute the only proton-attached centers to guide NMR analysis.

Figure 1.

(a) Original structures advanced for (−)-lomaiviticins A, B, and C (1a, 2a, and 3a, respectively), and the structures of (−)-kinamycin F (4), the synthetic hydroxyfulvene 5, and the aglycon 6a, which was first prepared in 2011. (b) Revised structures of (−)-lomaiviticins A, B, and C (1b, 2b, and 3b, respectively) based on the data reported herein. The assignments that are exchanged in the revised structures are shown in orange in 1b. (c) The structural homology between (−)-lomaiviticins A–C has been established by semisynthesis.^5,6

The structures 1a and 2a were based on NOESY, HMQC, and HMBC analysis. The data points from the initial report² that are germane to the work below are as follows. H2 and H4 were observed as singlets in the ¹H NMR spectra (500 MHz) of (−)-lomaiviticins A and B but were correlated in the COSY spectrum of (−)-lomaiviticin A. This correlation was attributed to W-plane coupling, thus forming the basis for assignment of a syn-facial arrangement of the C2 and C4 hydrogens, and the number of bonds separating them (four). A weak HMBC correlation between H4 and the diazo carbon (C5) in (−)-lomaiviticin A was reportedly observed; this correlation was attributed to 3J_C,H coupling and used to anchor the location of the aminosugar with respect to the diazocyclopentadiene. Finally, a correlation between H2 and C2 was observed in both the HMQC and HMBC spectra of (−)-lomaiviticin A. These correlations supported assignment of the location of the bridging carbon–carbon bond as that shown. While the biosynthetic relationship between (−)-lomaiviticins A and B has not been rigorously established, the 2,6-dideoxy α-l-oleandrose residues of (−)-lomaiviticin A are acid-labile. Hydrolytic cleavage of these residues, followed by 2-fold hemiketal formation, converts natural (−)-lomaiviticin A to semisynthetic (−)-lomaiviticin B (Figure 1c).⁵

In 2012, the Herzon laboratory reported the isolation of (−)-lomaiviticin C, the product of formal hydrodediazotization of (−)-lomaiviticin A.⁶ The C₁-symmetric structure 3a was assigned to (−)-lomaiviticin C based upon the doubling of many signals in the NMR spectra, a reduction in 26 mass units relative to (−)-lomaiviticin A, and the successful conversion of natural (−)-lomaiviticin C to semisynthetic (−)-lomaiviticin A (Figure 1c). New signals observed at 6.72 and 120.9 ppm in the ¹H and ¹³C NMR spectra, respectively, of (−)-lomaiviticin C were attributed to the hydroxfulvene shown in red. These shifts are in agreement with fully synthetic hydroxyfulvenes, such as 5 (δ H = 6.74, δ C = 120.0). Acidic digestion of (−)-lomaiviticin C allowed for isolation of the carbohydrate residues. Their absolute stereochemistry was determined by comparison of optical rotation data to synthetic standards (both were found to be of the l-form). ROESY analysis was then used to relay this assignment to the cyclohexane core. However, to the best of our knowledge, single crystal X-ray crystallographic data has not been obtained for any lomaiviticin. Moreover, none of the structures 1a–3a have yet been prepared by chemical synthesis.^5,7 The Herzon laboratory reported a synthesis of the aglycon structure 6a in 2011.^5,7j Unfortunately, attempted hydrolytic cleavage of the carbohydrate residues in natural (−)-lomaiviticin A could not be effected without accompanying decomposition of the aglycon.^5,8 Thus, a definitive link between fully synthetic and natural lomaiviticins has not been established in the 20 years since their initial disclosure.

Here, we report microED analysis of (−)-lomaiviticin C. This analysis indicates the structure of (−)-lomaiviticin C is that shown as 3b (Figure 1b). In this structure, the C1/C4 and C2/C3 positions are exchanged, and the configuration of C4 is inverted, relative to 3a. Since structural homology between (−)-lomaiviticins A, B, and C has been definitively established (vide supra, see also Figure 1c) these findings indicate the structures of (−)-lomaiviticins A and B are correctly depicted as 1b and 2b, respectively. High-field (800 MHz) NMR analysis of (−)-lomaiviticin C supports the structure 3b and indicates that earlier efforts may have been misled by a fortuitous combination of a near-zero ³J_H4,H5 coupling constant and a ⁴J_H4,C5 HMBC correlation that was interpreted as a ³J_H4,C5 coupling.

RESULTS

We chose (−)-lomaiviticin C for study because the isolate is more abundant (liquid fermentation yields of (−)-lomaiviticins A and C are ~0.5–1 and 37–62 mg per liter, respectively),⁶ it is more stable than (−)-lomaiviticin A,⁹ and it is >2 orders of magnitude less cytotoxic.⁶ Additionally, the C₁-symmetry of (−)-lomaiviticin C provides a nearly 2-fold increase in the number of NMR resonances and allows for detection of correlations between the two halves of the isolate, thereby greatly amplifying the utility of NMR analysis. We envisioned that microED, an emerging cryoEM method that allows for structural characterization of complex molecules from crystals that fall below the size required for common crystallographic techniques,¹⁰ might be uniquely effective in providing crystallographic data for (−)-lomaiviticin C.

In our initial studies, we found that deposition of 2.0 μL of a solution of (−)-lomaiviticin C (~200 ng) in tetrahydrofuran–acetonitrile (~0.1 mg/mL) onto a carbon TEM grid, followed by room temperature evaporation, yielded microcrystalline domains that diffracted to >2 Å. While these were insufficient for ab initio structural determination, we were encouraged by these initial findings. Ultimately, after extensive screening of sample preparation conditions, it was found that slow evaporation (30 d) of this solution at −20 °C led to microcrystals that diffracted to 1.05 Å (complete crystallographic parameters are presented in Table S1). Cryogenic data collection was required to mitigate radiation damage during collection of rotation data. Merging data from four crystals afforded an ab initio solution (R1 = 11.9%, refined anisotropically; Figure 2; see also Figure S1). From the preliminary structure, which was obtained without any human input other than the molecular formula, it was apparent that there was extensive structural variation from previous reports that relied exclusively on NMR analysis. Notably, the C1/C4 (C1′/C4′) and C2/C3 (C2′/C3′) positions were exchanged, and the configurations of C4 and C4′ were inverted. The connectivity of the atoms was determined by microED analysis. However, the positions of the heteroatoms in the aminosugar residues could not be assigned using the microED data alone. These heteroatoms were manually introduced to the locations established by earlier chemical degradation studies.⁶ The experimental and calculated powder diffraction data for structure 3b were in agreement (Figure S2).

Figure 2.

Preliminary (left) and fully refined (right) microED structures of (−)-lomaiviticin C (3b).

In light of these findings, we acquired high-field (800 MHz) NMR spectroscopic data for (−)-lomaiviticin C in acetone-d₆, which was found to provide the optimal balance of solubility and consistency of shimming among a number of solvents examined. A Monte Carlo molecular dynamics simulation¹¹ was carried out on structures 3a and 3b; the lowest energy conformers are shown in Figure 3. Notably, the calculated geometry of the lowest-energy conformer of 3b was in qualitative agreement with the geometry obtained by microED analysis (compare Figures 2 and 3). Additionally, the lowest-energy conformer of 3a contains an equatorial disposition of H2 and H4, which is consistent with the conformational analysis in the original structural studies.²

Figure 3.

Lowest-energy conformers (top) and line structures (bottom) of structures 3a and 3b. For clarity, the naphthoquinone residues are shaded in the lowest-energy conformers and omitted from the line structures.

NMR signals that were used to distinguish between the structures 3a and 3b are discussed below and presented in Figure 4 (for complete data, see Table S2). ¹H NMR analysis revealed H2 and H2′ each as a doublet of doublets at δ 3.95 (J = 4.2, 1.5 Hz) and δ 3.85 (J = 4.1, 1.3 Hz), respectively. These were clearly coupled to each other and to H4 and H4′ in the COSY spectrum. H4 and H4′ were observed as narrow doublets at δ 5.51 (J = 1.1 Hz) and δ 5.33 (J = 1.3 Hz), respectively. A TOCSY experiment showed correlations between (H2, H4′) and (H2′, H4), thereby confirming these protons are part of the same spin system (see the Supporting Information).

Figure 4.

Diagnostic ROESY and HMBC correlations in the NMR spectra of (−)-lomaiviticin C supporting structure 3b. (a) ROESY correlations between (H4, H2), (H4′, H2′), (H4, H2′), and (H4′, H2). (b) ROESY correlation between (H4, H5′). (c) HMBC correlations between (H4, C2′), (H4, C2), (H4′, C2′), and (H4′, C2). (d) HMBC correlations between (H12, C1) and (H12′, C1′).

2D-ROESY analysis supported the assignment 3b but not 3a. We observed a clear correlation between (H2, H4) and (H2′, H4′) (Figure 4a). These correlations are consistent with structure 3b wherein the pairs of protons are vicinal to each other. The structure 3a necessitates that H2, H4, H2′, and H4′ are equatorially disposed to account for the putative W-plane coupling discussed above. However, this spatial arrangement would not be expected to give rise to the ROESY correlation shown in Figure 4a. Figure 4a also shows weaker but definitive correlations between (H2′, H4) and (H2, H4′). These correlations across the monomeric units are accommodated by the structure 3b, but would not be anticipated in structure 3a, wherein the pairs of protons are oriented on opposing sides of each cyclohexane ring. We also observed a ROESY correlation between (H5′, H4) (Figure 4b). This is readily accommodated by the position and configuration of H4 in 3b but not the alternate configuration found in structure 3a. Additional ROESY correlations support the revised structure assignment (see the Supporting Information).

HMBC analysis (using a ⁿJ_C,H threshold of 8.0 Hz) provided further support for the structure 3b. We observed correlations between (H4, C2) and (H4′, C2′), which are consistent with either structure 3a or 3b (³J_C,H coupling in 3a, ²J_C,H coupling in 3b; Figure 4c). However, we detected additional correlations between (H4, C2′) and (H4′, C2). These latter correlations require ³J_C,H coupling in structure 3b but would necessitate a ⁴J_C,H coupling in structure 3a. However, a ⁴J_C,H coupling is inconsistent with the ROESY data shown in Figure 4a. Finally, correlations between (H12, C1) and (H12′, C1′) were observed (Figure 4d). Interpretation of these correlations as ³J_C,H couplings support the location of the carbonyl as that shown in 3b but would necessitate invoking a ⁴J_C,H coupling through an aliphatic system to achieve consistency with structure 3a. On reducing the ⁿJ_C,H threshold to 3.0 Hz, we detected a faint HMBC correlation between (H4, C5), as noted in the original characterization of (−)-lomaiviticin A,² but not between (H4′, C5′) or (H5′, C4′) (see the Supporting Information).

To provide further support for the structure revision, we carried out computational modeling of ¹³C NMR chemical shifts according to the method of Hehre et al.¹² We focused on lomaiviticin aglycon and (−)-lomaiviticin B because their caged structures significantly decrease the conformational flexibility of the molecules, thereby reducing the computational cost. To benchmark the computational approach, we first calculated the expected 13C NMR chemical shifts of the aglycon structure 6a (shown in Figure 1a), for which experimental 13C shifts have been recorded.^5,7j Figure 5a depicts the absolute difference between the experimental and calculated chemical shifts as a function of position. We observed good agreement between the calculations and experiment, with all but three positions (C4, C11, and C11a) within 5 ppm of experimental values, and an overall root-mean-square (RMS) = 3.39. These results support this computational approach as a viable strategy to determine the expected ¹³C NMR spectra of the lomaiviticins.

Figure 5.

(a) Absolute difference between calculated and experimental ¹³C chemical shifts of the aglycon structure 6a. RMS = 3.39. (b) Absolute difference between experimental ¹³C chemical shifts of the aglycon structure 6a and natural (−)-lomaiviticin B. RMS = 5.82. (c) Absolute difference between calculated and experimental ¹³C chemical shifts of the aglycon structure 6b and (−)-lomaiviticin B, respectively. RMS = 1.99.

We then compared the experimental ¹³C chemical shifts of the aglycon structure 6a with those of natural (−)-lomaiviticin B (Figure 5b). The experimental shifts were in poor agreement (RMS = 5.82). A 17.9 ppm difference was observed between the bridging carbon atoms (C2), and large differences were also observed at the all-substituted C3 stereocenter (Δδ = 10.6 ppm) and the center bearing the aminosugar residue (C4; Δδ = 9.6 ppm). We then determined the expected ¹³C chemical shifts of the revised aglycon structure 6b and compared these to the experimental chemical shifts of natural (−)-lomaiviticin B (Figure 5c). Among the three data sets shown in Figure 5, these data were in closest agreement, with an overall RMS = 1.99. All experimental chemical shifts of (−)-lomaiviticin B were within 5 ppm of the values calculated for the structure 6b.

We also calculated the expected ¹³C NMR chemical shifts of the structures 2a and 2b and compared these to the experimental chemical shifts of natural (−)-lomaiviticin B (Figure 6). With the exception of C4, all chemical shifts calculated for 2b were within 5 ppm of the corresponding shifts of natural (−)-lomaiviticin B (RMS = 2.83). By comparison, the structure 2a showed large discrepancies in the critical C2, C3, and C4 positions of the cyclohexene ring, at two adjacent carbon atoms (C11b, C12) and at the hemiketal carbon (C1; RMS = 6.63).

Figure 6.

Absolute difference between calculated ¹³C chemical shifts of structures 2a and 2b and natural (−)-lomaiviticin B. Gray series: 2a (RMS = 6.63); blue series: 2b (RMS = 2.83).

Finally, we measured ⁿJ_C,H coupling constants for H2, H2′, H4, and H4′ of natural (−)-lomaiviticin C using 2D-EXSIDE experiments. We determined the expected values of these couplings for structures 3a and 3b by DFT calculations. To render these calculations tractable, the single lowest energy conformers of 3a and 3b identified by molecular dynamics were used as input. These were each optimized by DFT (ωB97X-D/6–31G*; higher levels of theory do not lead to more accurate results, see ref 12) and the couplings were determined. Figure 7 depicts the absolute difference between the calculated and experimental coupling constants (because the 2D-EXSIDE experiment does not provide the sign of the coupling, the absolute values of the theoretical couplings were employed). As can be seen, the theoretical couplings derived from structure 3b were in better agreement (RMS = 1.80 vs RMS = 3.87 for 3a). The couplings between (H2, C4′) and (H2′, C4) are perhaps most diagnostic. The calculated ³J_C,H couplings for structure 3b were 6.3 and 6.1 Hz for (H2, C4′) and (H2′, C4), respectively, in reasonable agreement with the experimental values of 10.2 and 10.8 Hz. By comparison, structure 3a requires these to be ⁴J_C,H couplings, and we calculated ⁴J_C,H = 0 for both (H2, C4′) and (H2′, C4) in structure 3a. Additionally, the calculated ³J_H,H coupling constants for (H2, H4) and (H2′, H4′) in 3b were 1.6 and 1.1 Hz, respectively. These are in agreement with the experimental values of 1.1 and 1.3 Hz (vide supra). The calculated ⁴J_H,H couplings for structure 3a were 0.5 and 0 Hz for (H2, H4) and (H2′, C2′), respectively.

Figure 7.

Absolute difference between experimental ⁿJ_C,H coupling constants for H2, H2′, H4, and H4′ of natural (−)-lomaiviticin C and those expected for structures 3a and 3b based on DFT calculations. Experimental couplings were obtained by 2D-EXSIDE experiments, which do not provide the sign of the couplings. Therefore, the absolute values of the theoretical couplings were employed.

DISCUSSION

The lomaiviticins are complex cytotoxic bacterial metabolites that were first disclosed in 2001. They contain an unusual diazotetrahydrobenzo[b]fluorene functional group and the deoxyglycosides l-oleandrose and N,N-dimethyl-l-pyrrolosamine. The high degree of unsaturation of the lomaiviticins makes their de novo structure assignment exceedingly difficult using NMR analysis alone. In the 20 years since their initial disclosure, substantial efforts have been devoted toward their chemical synthesis. Yet, because none of the proposed structures 1a–3a have been synthesized, a comparison of spectroscopic data derived from fully synthetic material to that of the natural samples has not been possible. Moreover, single crystal X-ray crystallographic data of any lomaiviticin has not been disclosed.

Here, we used microED analysis of (−)-lomaiviticin C to revise the core structure of this family of metabolites. The results of the microED studies are supported by high-field NMR spectroscopy and computational modeling. The central issue at hand is the orientation of the cyclohexenone ring with respect to the diazofluorene residue. As best as we can discern from the original report, the connectivity shown in structures 1a and 2a rested on two observations. First, it was noted that a weak correlation was detected from H4 to C5 in the HMBC spectrum of (−)-lomaiviticin A (see Figure 8). At the time, this was quite reasonably interpreted as a ³J_C,H coupling, and this served to anchor the position of the secondary ether (bearing the aminosugar) with respect to the diazofluorene. In addition, both H2 and H4 were observed as singlets in the 500 MHz ¹H NMR spectrum of (−)-lomaiviticin A, but a COSY correlation was observed between these two protons. This COSY correlation was attributed to a syn diequatorial disposition of H2 and H4 (placing the aminosugar and bridging bonds in an axial orientation) giving rise to a 4-bond W-plane coupling. This coupling, in turn served to orient the bridging carbon atom (C2), both in position and stereochemistry, relative to C4, and the remaining substituents could be placed logically within the ring. It is worth noting that X-ray analysis of a 6,6′-bicyclohexenone structurally related to that found in (−)-lomaiviticin A revealed an axial disposition of the conjoining bond.¹³ In our analysis of natural (−)-lomaiviticin C, we did not observe an HMBC correlation from H4 to C5 using a ⁿJ_C,H threshold of 8 Hz; a faint coupling was observed when the ⁿJ_C,H threshold was reduced to 3 Hz. We attribute this correlation to a weak ⁴J_C,H coupling in the structure 3b, rather than a ³J_C,H coupling as originally proposed in structure 1a. Such four-bond couplings in unsaturated systems are well-precedented.¹⁴ Our molecular dynamics simulations indicate that the dihedral angles between (H2, H4) and (H2′, H4′) in (−)-lomaiviticin C are 78° and 86°, respectively. Consistent with this, the calculated ³J_H,H coupling constants for (H2, H4) and (H2′, H4′) in 3b were 1.6 and 1.1 Hz, respectively, which is in agreement with experiment (1.1–1.3 Hz). Such small couplings could give rise to an apparent singlet at lower frequency, as originally observed. In sum, the original assignment appears to have gone awry due to a limited number of H–H and C–H couplings deriving from the high degree of unsaturation of the isolates, and the combination of two unusual spin-coupling effects.

Figure 8.

Original (left) and revised (right) core structures of the lomaiviticins.

Because the interconversion of lomaiviticins A–C has been demonstrated (Figure 1c), the findings here apply to each of these metabolites. The oxidation pattern found in the revised structures 1b, 2b, and 3b has been identified in other isolates, such as nenestatin A (Figure 9),¹⁵ and in intermediates in kinamycin biosynthesis.^4c Interestingly, biosynthetic precursors to nenestatin A were found to undergo nucleophilic addition and dimerization,^15b a pathway that may have implications for lomaiviticin biosynthesis. The lomaiviticin biosynthetic gene cluster has been identified,¹⁶ but the mechanism of dimerization remains poorly understood.

Figure 9.

Structure of nenestatin A.

Studies from the Herzon laboratory have established that the cytotoxic effects of (−)-lomaiviticin A derive from the induction of double-strand breaks (DSBs) in DNA.¹⁷ Because these studies were conducted using semisynthetic samples of (−)-lomaiviticin A, derived from diazotransfer to natural (−)-lomaiviticin C, the conclusions of these studies remain valid. However, the mode of interaction of (−)-lomaiviticin A with DNA¹⁸ will need to be reinterpreted.

CONCLUSION

In conclusion, we have used a combination of microED, NMR, and computational analysis to advance revised structures for the lomaiviticins. In the revised structures 1b, 2b, and 3b the positions of the ketone and aminosugar (C1/C4), and the bridging carbon–carbon bond and the center bearing the oleandrose residues (C2/C3) are transposed. The configuration of the secondary ether bearing the aminosugar residue is also inverted. These studies highlight the power of the microED method to elucidate natural product structure in instances where NMR spectroscopy alone may be insufficient.