Structure Elucidation of Secondary Metabolites: Current Frontiers and Lingering Pitfalls

CONSPECTUS:

Analytical methods allow for the structure determination of submilligram quantities of complex secondary metabolites. This has been driven in large part by advances in NMR spectroscopic capabilities, including access to high-field magnets equipped with cryogenic probes. Experimental NMR spectroscopy may now be complemented by remarkably accurate carbon-13 NMR calculations using state-of-the-art DFT software packages. Additionally, microED analysis stands to have a profound effect on structure elucidation by providing X-ray-like images of microcrystalline samples of analytes. Nonetheless, lingering pitfalls in structure elucidation remain, particularly for isolates that are unstable or highly oxidized. In this Account, we discuss three projects from our laboratory that highlight nonoverlapping challenges to the field, with implications for chemical, synthetic, and mechanism of action studies. We first discuss the lomaiviticins, complex unsaturated polyketide natural products disclosed in 2001. The original structures were derived from NMR, HRMS, UV–vis, and IR analysis. Owing to the synthetic challenges presented by their structures and the absence of X-ray crystallographic data, the structure assignments remained untested for nearly two decades. In 2021, the Nelson group at Caltech carried out microED analysis of (−)-lomaiviticin C, leading to the startling discovery that the original structure assignment of the lomaiviticins was incorrect. Acquisition of higher-field (800 MHz ¹H, cold probe) NMR data as well as DFT calculations provided insights into the basis for the original misassignment and lent further support to the new structure identified by microED. Reanalysis of the 2001 data set reveals that the two structure assignments are nearly indistinguishable, underscoring the limitations of NMR-based characterization. We then discuss the structure elucidation of colibactin, a complex, nonisolable microbiome metabolite implicated in colorectal cancer. The colibactin biosynthetic gene cluster was detected in 2006, but owing to colibactin’s instability and low levels of production, it could not be isolated or characterized. We used a combination of chemical synthesis, mechanism of action studies, and biosynthetic analysis to identify the substructures in colibactin. These studies, coupled with isotope labeling and tandem MS analysis of colibactin-derived DNA interstrand cross-links, ultimately led to a structure assignment for the metabolite. We then discuss the ocimicides, plant secondary metabolites that were studied as agents against drug-resistant P. falciparum. We synthesized the core structure of the ocimicides and found significant discrepancies between our experimental NMR spectroscopic data and that reported for the natural products. We determined the theoretical carbon-13 NMR shifts for 32 diastereomers of the ocimicides. These studies indicated that a revision of the connectivity of the metabolites is likely needed. We end with some thoughts on the frontiers of secondary metabolite structure determination. As modern NMR computational methods are straightforward to execute, we advocate for their systematic use in validating the assignments of novel secondary metabolites.

Graphical Abstract

INTRODUCTION

Methods for the structure elucidation of complex secondary metabolites have advanced considerably. The availability of high-field NMR magnets, cryogenic probes, and advanced two-dimensional pulse sequences has positioned NMR spectroscopy as the primary mode of structure determination.⁵ Nonetheless, NMR-guided structure elucidation remains imperfect, especially for metabolites that are highly oxidized (e.g., low H:C ratio). Additionally, NMR-based structure elucidation presupposes that the metabolites of interest are stable to fractionation and are produced in quantities sufficient for isolation. Advances in genome-based approaches to metabolite discovery have made clear that many biosynthetic gene clusters (BGCs) are poorly expressed under laboratory conditions, if at all. Additionally, many secondary metabolites are unstable and not amenable to the multiple rounds of purification required by the classical “grind-and-find” approach.⁶

Here we discuss three projects involving the attempted synthesis and study of secondary metabolites for which the structure assignment was incorrect (lomaiviticins and ocimicides) or for which a structure assignment was not possible based on available data (colibactin). For two of these projects (colibactin and lomaiviticins), we show how we were able to overcome these obstacles to arrive at the correct structure assignment. The correct structure of the ocimicides remains unknown.

LOMAIVITICINS

In 2001, researchers at Wyeth Pharmaceuticals and the University of Utah described two cytotoxic dimeric metabolites from a strain of M. lomaivitiensis (later classified as S. pacifica).⁸ The more abundant isolate, (−)-lomaiviticin A, was advanced as structure 1 (Figure 1a). A second isolate, (−)-lomaiviticin B, was depicted as structure 2. The lomaiviticins possess diazotetrahydro[b]benzofluorene (diazofluorene) functional groups and two or four deoxyglycoside residues. Formally, structure 2 derives from the loss of the oleandrose residues of 1 and the addition of the resulting C3/C3′ hydroxyl substituents to the C1/C1′ carbonyl groups. (−)-Lomaiviticin A displayed antiproliferative activities in the low nM─pM range and antimicrobial minimum inhibitory concentrations (MICs) in the ng/spot range. The antiproliferative activity of (−)-lomaiviticin B was not evaluated (owing to its lower abundance), but it also displayed potent antimicrobial effects. The cytotoxic effects of (−)-lomaiviticin A originate from the induction of double-strand breaks in DNA.⁹

Figure 1.

a. Structures 1 and 2 were advanced for (−)-lomaiviticins A and B in 2001. b. Spectroscopic methods used to identify the substructures in (−)-lomaiviticins A and B. c. Core fragments identified by NMR analysis (¹³C, HMBC, and HSQC) and their assembly into structure 1. Key HMBC correlations are shown in orange and green (Note: due to the C₂ symmetry of (−)-lomaiviticin A, these correlations apply to both cyclohexenone rings but are depicted on separate rings for clarity.) d. Connectivity and relative configuration of the cyclohexenone derived from the bis(furanol) structure of (−)-lomaiviticin B and putative long-range (W-plane) coupling between H2 and H4 in (−)-lomaiviticin A. This correlation was assigned as a ⁴J_H2─H4 (W) coupling. A and B represent the oxygen-linked β-N,N-dimethyl-l-pyrrolosamine and α-l-oleandrose residues, respectively (see the text).

The structure elucidation of the lomaiviticins was difficult owing to their C₂ symmetry and unsaturation (Ω = 30–32).⁷ High-resolution mass spectrometry established the molecular formula of (−)-lomaiviticin A as C₆₈H₈₀N₆O₂₄. The observation of half the expected carbon and proton NMR signals pointed toward a C₂-symmetric structure. The connectivity and relative configuration relied on UV–vis and infrared spectroscopy and 1- and 2-dimensional NMR experiments (Figure 1b). The 5,8-dihydroxy-1,4-naphthoquinone was inferred based upon ¹H and ¹³C NMR analysis, HMBC correlations, and a UV absorption at 525 nm, which is characteristic of these structures. The diazo cyclopentadiene was supported by observation of a carbon-13 resonance at δ 78.8 ppm (diazo carbon atom, C5/5′) and a strong infrared stretch at 2148 cm⁻¹, diagnostic features of this functional group found in the monomeric diazofluorenes known as the kinamycins.¹⁰ The deoxyglycosides β-N,N-dimethylpyrrolosamine and α-oleandrose were identified by ¹H and ¹³C NMR, the analysis of ³J_H─H coupling constants, and a comparison to isolates containing these carbohydrate residues. The absolute configuration of either carbohydrate was not defined but was subsequently established as l by the degradation of (−)-lomaiviticin C (vide infra). The absolute configuration of the aglycon was assigned by analogy to the kinamycins.

Carbon-13, HSQC, and HMQC spectral data suggested that each cyclohexenone contained one carbonyl group, one ethyl substituent, two tertiary carbons, and three quaternary carbons (Figure 1c). A correlation between H2/H2′ and C2/C2′ in both the HMBC and HMQC spectra established the position of the cojoining carbon–carbon bond at C2. The location of the oleandrose residue was assigned by ²J_H2─C3 and ²J_H4─C3 couplings. A weak HMBC correlation between H4/H4′ and C5/C5′ (diazo carbon) was interpreted as a ³J_C─H coupling, anchoring the orientation of the cyclohexenone ring relative to the diazofluorene residue.

The structure elucidation of (−)-lomaiviticin B established the relative configuration of the cyclohexenone in both isolates. The fused furanol structure 2 was logically advanced based on the absence of the oleandrose and ketone residues (HRMS and NMR) in (−)-lomaiviticin A (Figure 1a and d). It is plausible that (−)-lomaiviticin B derives from the 2-fold deglycosylation of (−)-lomaiviticin A, followed by cyclization. This structure necessitates a cis-disposition of the dimeric bond and C3 oxygen substituents. This stereochemical assignment was extended to (−)-lomaiviticin A based on the assumption that the configuration of the cyclohexenone core is identical in the two isolates. Consistent with this, both H2 (H2′) and H4 (H4′) appear as singlets in the 500 MHz ¹H NMR spectrum of (−)-lomaiviticin A. However, a correlation between these protons was observed in the COSY spectrum and was attributed to a four-bond W-coupling, where ⁴J_H─H < 2 Hz is typical.¹¹ This putative W-coupling requires a cis-diequatorial disposition of H2 (H2′) and H4 (H4′), supporting the same relative stereochemical arrangement for both isolates (Figure 1d).

From 2012 to 2013, (−)-lomaiviticin C was isolated by our laboratory¹² and by Moore and co-workers (Figure 2a).¹³ NMR and HRMS analysis revealed that (−)-lomaiviticin C is nearly identical to (−)-lomaiviticin A, leading to the assignment of structure 3 (Figure 2a). The only difference between structures 1 and 3 is the presence of a hydroxyfulvene in (−)-lomaiviticin C (orange in Figure 2a), as opposed to two diazofluorenes in (−)-lomaiviticin A. The hydroxyfulvene was supported by features at δ 6.72 ppm (H5′, s, 1H) and δ 120.9 ppm (C5′) in the H-1 and C-13 NMR spectra, respectively, of (−)-lomaiviticin C, which are similar to synthetic hydroxyfulvenes.¹⁴ Doubling of most signals (relative to (−)-lomaiviticin A) was observed, supporting a C₁-symmetric structure.

Figure 2.

a. Structure 3 advanced for (−)-lomaiviticin C in 2012. Acid-catalyzed hydrolysis of the carbohydrate residues followed by optical analysis established that both possess the l configuration. b. ROESY correlations used to elucidate the absolute configuration of the aglycon in the lomaiviticins are shown in orange and green. A and B represent the oxygen-linked β-N,N-dimethyl-l-pyrrolosamine and α-l-oleandrose residues, respectively.

Acidic digestion of (−)-lomaiviticin C allowed for isolation of the carbohydrate residues; optical analysis established that both are of the l-form. The configuration of the aminosugar was used to infer the absolute configuration of the aglycon (Figure 2b). A ROESY correlation between H1A and H4 requires a cis-diaxial disposition of these protons relative to the C─O─C plane. An additional ROESY correlation between H2A_eq and H12 is then accommodated only by the diastereomer shown, since it requires both protons to be on the same face of the cyclohexenone ring. Diazotransfer to (−)-lomaiviticin C provided semisynthetic (−)-lomaiviticin A,¹² and hydrolysis of the oleandrose residues of semisynthetic (−)-lomaiviticin A provided semisynthetic (−)-lomaiviticin B.¹⁵ Semisynthetic samples of (−)-lomaiviticins A and B obtained in this way were indistinguishable from natural material, establishing the structural homology among the three isolates.

In 2019, we were contacted by Prof. Hosea Nelson and co-workers, who were interested in obtaining complex isolates for microED analysis. Initially developed for elucidating the structures of biological macromolecules as frozen hydrated crystals, microED analysis is rapidly emerging as a powerful method for the structure determination of small organic molecules using nanocrystalline material (~10⁻¹⁵ g).¹⁶ We provided a sample of (−)-lomaiviticin C fully expecting that the microED analysis would be confirmatory. A preliminary structure obtained in early 2021¹⁷ led to the startling discovery that the original assignment of the lomaiviticins was incorrect.⁴

The microED study (Figure 3a) revealed that the C1 and C4 positions were exchanged and the configuration of C4 was inverted relative to the original assignment. Since the homology among (—)-lomaiviticins A, B, and C was established (vide supra), the structures of each isolate were revised as shown in Figure 3b.

Figure 3.

a. MicroED structure of (−)-lomaiviticin C. b. Revised structures of the lomaiviticins. c. Absolute difference between theoretical carbon-13 chemical shifts of structures 2 (see Figure 1a) and 7 and natural (−)-lomaiviticin B. Carbon-13 shifts were determined from the Boltzmann average of the theoretical carbon-13 shifts of the three low-energy conformers of 2 and 7 (ωB97X-V/6-311+G(2DF,2P)[6-311G*]). Gray series: 2 (root-mean-square error (RMSE) = 6.63); blue series: 7 (RMSE = 2.83).

DFT calculations¹⁸ were conducted to compare the theoretical carbon-13 shifts of the original (2) and new (7) (−)-lomaiviticin B structures to experimental values of the natural isolate. (−)-Lomaiviticin B was selected due to the presence of fewer carbohydrate residues and a more rigid structure. The root-mean square errors (RMSEs) between the theoretical and experimental carbon-13 shifts of (−)-lomaiviticin B were 6.63 and 2.83 for structures 2 and 7, respectively (Figure 3c). The theoretical C─H and H─H coupling constants for the new structure of (─)-lomaiviticin C (8) were also in better agreement with the experimental values.

We carried out high-field NMR studies (800 MHz ¹H, 200 MHz ¹³C, cold probe) of natural (−)-lomaiviticin C to probe the revised structure and gain insight into where the original structure determination went awry. We observed many signals consistent with the revised structure but inconsistent with the original assignment. A selection of diagnostic HMBC and ROSEY correlations is shown in Figure 4. HMBC correlations observed between H4 and C2′ (and H4′ and C2) and H12 and C1 (and H12′ and C1′) would require ⁴J_C─H couplings through an aliphatic system in the original structure 3, which are rare (Figure 4a).¹⁹ In the revised structure 8, these correlations are assigned as common ³J_C─H couplings. Molecular dynamics simulations were used to determine the lowest-energy conformers of 3 and 8 (Figures 2a and 3b). The lowest-energy conformation of 8 was in qualitative agreement with the microED structure (Figure 3a). These structures were then used to guide ROESY analysis (Figure 4b). A ROESY correlation observed between H2 and H4 (and H2′ and H4′) seems implausible in structure 3, as it would require a ring inversion to place H2/H4 or H2′/H4′ in pseudoaxial positions. In structure 8, H2 and H4 (and H2′ and H4′) occupy a vicinal, trans-diequatorial relationship, and a ROESY correlation would be expected. Not depicted in Figure 4b (for clarity) are weak ROESY correlations between H2 and H4′ (and H2′ and H4). These would be unlikely in structure 3, as the protons are located on opposite faces of each cyclohexenone ring. However, these correlations can be interpreted as a transannular interaction arising from the cis-disposition of each pair of protons (with respect to each cyclohexenone ring) in structure 8.

Figure 4.

a. Selected HMBC correlations observed for (−)-lomaiviticin C in 2021 (800 MHz ¹H, cold probe) and their application to structures 3 and 8. b. Selected ROESY correlations observed for (−)-lomaiviticin C in 2021 (800 MHz ¹H, cold probe) and their application to structures 3 and 8. c. Reported HMBC correlations observed for (−) lomaiviticin A in 2001 (500 MHz ¹H) and their application to structures 1 and 6. d. Additional HMBC correlations observed for (−)-lomaiviticin C in 2021 (800 MHz ¹H, cold probe) and their application to structures 3 and 8. A and B in Figure 4a,b represent the oxygen-linked β-N,N-dimethyl-l-pyrrolosamine and α-l-oleandrose residues, respectively.

Though we do not have access to the NMR data files from the 2001 study (and the Wyeth team has, to the detriment of the field, been disbanded), an inspection of tabulated and graphical NMR spectra suggests that the two structural series discussed above are essentially indistinguishable using the 2001 data. Figure 4c depicts the HMBC correlations to H4 in (−)-lomaiviticin A available in this data set (except for C1A). All of these correlations correspond to common ²J_C─H and ³J_C─H couplings in structure 1. We detected a weak HMBC correlation between H4 and the diazo carbon (C5) in our 2021 studies of (−)-lomaiviticin C. In the revised structure of (−)-lomaiviticins A and C (6 and 8, respectively), this correlation is assigned as a well-precedented allylic ⁴J_C─H coupling.²⁰ Additionally, the HMBC spectrum of (−)-lomaiviticin C obtained in 2021 revealed new correlations between H4 and C2′, H4 and C1, and H4 and C11a (Figure 4d). The correlations from H4 to C11a and H4 to C1 are theoretically accommodated by both 3 and 8 (4-bond allylic couplings in 3; 3-bond and 4-bond allylic couplings in 8), but the correlation from H4 to C2’ occurs through an aliphatic system and would be expected only in 8. This correlation is present only in C₁-symmetric (−)-lomaiviticin C, as the C₂-symmetric structure of (−)-lomaiviticin A makes this indistinguishable from a ³J_H4─C2 coupling. This signal was overlooked in the original isolation of (−)-lomaiviticin C due to the inability to clearly resolve C2 and C2′ on a 500 MHz instrument.

COLIBACTIN

The intestine contains diverse microorganisms that impact physiology and disease.²¹ Numerous correlations between microbiome composition and host physiology are documented, but the isolation and characterization of metabolites that may underpin these correlations are challenging. Such characterization is required if researchers are to probe for metabolite-dependent causal relationships between the microbiome and human disease.

In 2006, it was discovered that certain E. coli and other proteobacteria harbor a 54-kb BGC termed clb (aka pks) which encodes the biosynthesis of the genotoxin colibactin.²² Mammalian cells exposed to clb⁺ E. coli accumulated DNA double-strand breaks (DSBs). Mice infected with clb⁺ E. coli developed tumors under inflammatory conditions, and clinical data revealed an increased prevalence of clb⁺ bacteria in colorectal cancer (CRC) patients.²³ Despite the strong evidence supporting a role for colibactin in CRC, our inability to isolate the metabolite and elucidate its structure hampered the study of its role in carcinogenesis. In our minds, this points to an important frontier in the field: Advances in genomics provide methods to rapidly identify novel BGCs, but how does one pursue the characterization of these metabolites when they are unstable or produced in quantities too small for isolation? Certainly, this challenge has been recognized by the community.²⁴ Here we describe how we used a multidisciplinary approach to elucidate the structure and mechanism of action of colibactin. For a more detailed discussion, the reader is directed to ref 25.

The clb gene cluster encodes an inactive biosynthetic intermediate termed precolibactin, which contains an N-myrisotyl-d-Asn residue.²⁶ Precolibactin is transported to the periplasm²⁷ where it is converted to colibactin by removal of the N-myrisotyl-d-Asn residue by the serine protease ClbP.²⁶ As researchers were unable to isolate colibactin from wild-type bacteria possessing an intact clb pathway, a strategy involving the large-scale fermentation of genetically modified clb⁺ E. coli strains was pursued.^26c These studies primarily employed clbP mutant strains, based on the underlying hypothesis that ClbP inactivation would result in the accumulation of precolibactin. These clbP mutant strains were often further modified by the knockdown of other genes. This approach has led to the identification of >40 clb products to date.²⁸ These isolates have provided substantial insight into colibactin’s structure and the function(s) of the modified enzymes.²⁹ However, while the mutation of clbP was thought to simply promote the accumulation of stable clb products, we discovered that this genetic modification derails the biosynthetic pathway.³⁰ Herein, we focus on the clb products precolibactin C (9), precolibactin 886 (10), colibactin 770 (11), and colibactin 771 (12, Figure 5). We outline how the study of these metabolites contributed to our understanding of colibactin’s biosynthesis, its molecular mechanism of action, and the structure of colibactin 771 (12) itself, which is thought to be the final clb product but which has still eluded direct isolation. Recently, a degradation product of colibactin 772 (12) was isolated from wild-type clb⁺ E. coli.³¹

Figure 5.

Structures of selected products derived from mutant (for 9 and 10) or wild-type (for 11 and 12) clb⁺E. coli.

Precolibactin C (9) was predicted³² in 2015 and later isolated from a mutant clb⁺ E. coli strain (Figure 5).³³ Biosynthetic studies indicated that precolibactin C (9) was off-loaded from the assembly line as its linear precursor before undergoing nonenzymatic cyclization to generate the pyridone residue.³⁴ The presence of a cyclopropane led to speculation that colibactin’s genotoxicity results from the alkylation of DNA by ring-opening addition,³⁵ a mechanism of action established for other classes of metabolites.³⁶ However, we posited that these pyridones would be poor DNA alkylating agents, as cyclopropane opening would require disruption of aromaticity. We hypothesized that these isolates are biosynthetic derailment products derived from the persistence of the N-myrisotyl-d-Asn residue (Figure 6). We proposed that in wild-type strains, ClbP-mediated deacylation of linear precolibactins, such as 13, would trigger an alternative pathway to generate an unsaturated imine residue (14), a scaffold that had been previously proposed³⁵ but which had eluded isolation.

Figure 6.

In clbP mutant clb⁺ E. coli, the linear biosynthetic intermediate 13 undergoes a 2-fold cyclodehydration to yield nongenotoxic pyridones, such as precolibactin C (9). Synthetic studies revealed that in wild-type clb⁺ E. coli, ClbP-mediated deacylation of linear precolibactins, such as 13, provides unsaturated imines, such as 14, which are potent DNA alkylating agents.

We used a chemical synthetic approach to show that both cyclization pathways were viable from the same intermediate, providing access to compounds containing either pyridone or unsaturated imine residues.³⁷ Mechanism-of-action studies established that the unsaturated imines, but not pyridine-containing analogs, alkylate DNA by ring-opening addition.³⁰ This model was further supported by the observation that synthetic unsaturated imines such as 14 are substrates of the resistance enzyme ClbS, a cyclopropane hydrolase that protects clb⁺ bacteria from autotoxicity.³⁸ The mutation of clbP promotes the accumulation of metabolites containing N-myrisytoyl-d-Asn, but the persistence of this residue diverts the biosynthetic pathway toward unnatural, pyridone-containing products such as precolibactin C (9).

Efforts to characterize more advance precolibactin metabolites resulted in the isolation of the macrocyclic precolibactin 886 (10; Figure 5).³⁹ Precolibactin 886 (10) is one of the most biosynthetically advanced precolibactins, with only three clb enzymes unaccounted for in its biosynthesis. The unusual structure of precolibactin 886 (10) and a low titer (2.8 mg isolated from a 1000 L fermentation) raised questions about its origin. We synthesized ketoimine 15, the putative precursor to precolibactin 886 (10, Figure 7).² All attempts to achieve the macrocyclization of 15 were unsuccessful. Instead, products resulting from the hydrolysis of the ketoimine and, surprisingly, cleavage of the C36─C37 bond were formed. In light of this, we began to suspect that the cyclization may occur during analytical or preparative HPLC, which had been used to both monitor the cultures and isolate precolibactin 886 (10).³⁹ Subjecting 15 to semipreparative HPLC purification provided precolibactin 886 (10; 3%).² The low mass balance likely derives from decomposition by C36─C37 bond cleavage during the purification (discussed above). Our findings make clear that the oxidized two-carbon spacer renders advanced clb products unstable. We cannot rigorously exclude the possibility that macrocyclization does occur during fermentation (and we were simply unable to recapitulate these conditions in the laboratory), but our studies would seem to suggest that macrocyclic precolibactins are artifacts generated during the analytical and purification processes. A more complex derivative of precolibactin 886 (10) was subsequently isolated.⁴⁰ This isolate was not completely characterized, and the structure assignment is likely incorrect.⁴¹

Figure 7.

Studies suggest that the macrocyclic precolibactin 886 (10) is an artifact resulting from the cyclization of a linear ketoimine precursor during the analytical and purification processes. Advanced precolibactins containing the ketoimine residue are susceptible to C36─C37 bond cleavage.

While the isolation studies discussed above provided advanced precolibactins and the mechanism-of-action studies provided the first link between metabolite structure and clb genotoxicity, the structure assignment of colibactin remained incomplete. In 2018, Nougayrède and co-workers disclosed that HeLa cells infected with clb⁺ E. coli accumulated DNA interstrand cross-links (ICLs).⁴² At first glance, this observation appears to be inconsistent with the DSB phenotype reported earlier.²² However, the transient production of DSBs is an obligate step in the repair of ICLs by the Fanconi anemia (FA) pathway.^41a Separately, it was known that all of the biosynthetic enzymes in the clb gene cluster are required for the genotoxic phenotype.^22,42 Collectively, these observations suggested to us that the vestiges of colibactin may be entrained in the ICLs first observed by Nougayrède. Characterization of the colibactin-derived ICL might provide a means to infer its structure.

To achieve this, we carried out high-resolution LC-MS/MS analysis of DNA–colibactin adducts arising from the digestion of exogenous DNA that had been added to cultures of wild-type and auxotrophic clb⁺ E. coli (the latter were supplemented with carbon-13-labeled amino acids).^43,3 The structure of the bis(adenine) adduct 16 was established by this approach (Figure 8). Though the location of the adenine base could be confidently assigned, we could not assign the site of adenine alkylation. The N3-linked structure shown is based on studies by Balskus and co-workers, who established that synthetic colibactin fragments^44,37 form monoadenine adducts linked through N3.⁴

Figure 8.

Isotope labeling and tandem MS analysis of clb⁺ E. coli induced-DNA ICLs led to the identification of colibactin–DNA adduct 16. α-Aminoketone colibactin 771 (12) is proposed as the structure of colibactin based on biosynthetic logic. Colibactin 770 (11), which arises from aerobic oxidation and the hydrolysis of 12, was observed in bacterial extracts; its structure was confirmed by chemical synthesis. DNA ICLs induced by synthetic samples of 11 were indistinguishable (by tandem MS analysis) from those induced by clb⁺ E. coli.

The structure 16 was further supported by the observation of two mono(adenine) adducts arising from oxidative hydrolysis of the C36─C37 bond of 16 (Figure 7). Based on biosynthetic logic and the analysis of the DNA adduct 16, we proposed the α-aminoketone 12 (colibactin 771) as the structure of colibactin. While undetectable in bacterial extracts, we did observe and characterize the diketone colibactin 770 (11), which derives from aerobic oxidation and hydrolysis of the α-aminoketone residue of 12. DNA adduct 16 arises from opening of the cyclopropane of 11 or 12 by adenine. We synthesized colibactin 770 (11) and demonstrated that it was identical to natural material by LC/MS co-injection. DNA cross-links induced by synthetic colibactin 770 (11) were indistinguishable (by tandem MS analysis) from those induced by clb⁺ E. coll.³ A similar structure assignment was advanced contemporaneously by Balskus and co-workers.⁴⁶ Notably, the bis(adenine) adduct 16 was observed as its doubly charged cation in the MS spectra of DNA digests. The singly charged cation was ~10³ less abundant, as expected for a small molecule that contains two basic residues.

OCIMICIDES

In 2010, the structures and antimalarial activities of ocimicide A₁ (17), ocimicide B₁ (18), and ocimicide C₁ (19) and semisynthetic derivatives (+)-ocimicide A₂ (20), (−)-ocimicide B₂ (21), (−)-ocimicide C₂ (22), and (+)-ocimicide C₃ (23) were disclosed in the patent literature (Figure 9).⁴⁷ The natural ocimicides were obtained from extracts of Ocimum sanctum root bark. The ocimicides reportedly demonstrated nM antimalarial activity against chloroquine-sensitive and -resistant strains of Plasmodium falciparum and high selectivities toward malarial parasites over adult human epithelial cells. The semisynthetic derivatives 20–23 provided mice with prophylactic protection from Plasmodium berghei infection⁴⁷ and effected a radical cure in rhesus monkeys,^47b without detectable toxicity.

Figure 9.

Reported structures of a. ocimicide A₁ (17), ocimicide B₁ (18), and ocimicide C₁ (19) and b. semisynthetic derivatives (+)-ocimicide A₂ (20), (−)-ocimicide B₂ (21), (−)-ocimicide C₂ (22), and (+)-ocimicide C₃ (23). The donor–acceptor cyclopropane system is highlighted in blue in 17.

The ocimicide alkaloids (17–23) were reported to contain a rigid hexacyclic core comprising a tetrasubstituted aminocyclopropane, a pyrrolidine ring, and a quinoline ring with apparently alternating orientations (compare 17 and 18). Notably, the central cyclopropane ring is part of a reactive donor–acceptor system.⁴⁸ The structures of 17–23 were established by HRMS, IR, UV–vis, and NMR, but no data was provided for the natural isolates.⁴⁷ Tabulated spectroscopic data were reported for the semisynthetic derivatives 20–23.

We developed a synthetic approach to ocimicide A₁ (17) that employed a stereochemical relay to establish the azabicyclo[3.1.0]hexane core, which contains a tetrasubstituted cyclopropane (Figure 10).¹ The alkene within 24 (prepared in five steps and 56% yield from 4-methoxypyridine) was cleaved and converted to a carboxylic acid. Stereoselective bromolactonization generated the bromolactone 25. Exposure of the unpurified bromolactone 25 to potassium carbonate in methanol provided the epoxy ester 26 (45% from 24). Deprotonation of 26, followed by heating, induced an epoxide-opening–ring-contraction reaction to establish the azabicyclo[3.1.0]hexane 27 (44%). Mesylation of 27, followed by treatment with sodium methoxide, provided the methyl imidate 29. Acid-catalyzed hydrolysis of 29 then generated the lactam 30 (96% from 27). By this approach, the relative configuration of the azabicyclo[3.1.0]hexane residue was derived from the bromolactonization step. Saponification of the methyl ester 30, followed by Weinreb amide formation, provided the crystalline Weinreb amide 31, whose structure was confirmed by X-ray analysis.

Figure 10.

Synthesis of lactam 30 and Weinreb amide 31. To track the stereochemical relay, newly formed bonds in each step are highlighted in blue. The ³J_H─H coupling constants between the protons shown in orange in 30 did not agree with those reported for ocimicide A₁ (17).

The tert-butyl carbamate of 31 could be removed under acidic conditions, and the resulting ammonium salt 32 was stable (Figure 11). However, all attempts to form the free base of 32 resulted in uncharacterized decomposition products. It seemed plausible that ring opening of the donor-acceptor cyclopropane occurred (inset of Figure 11). The ocimicides themselves were reportedly stable isolates.⁴⁷ Analysis of the coupling constants within the azabicyclo[3.1.0]hexane core revealed inconsistencies between our intermediate 30 (Figure 10) and the natural and semisynthetic isolates. The ³J_H─ coupling constants within 30 (5.2 and 0 Hz) did not match those reported for (+)-ocimicide A₂ (20; 9.2 and 4.4 Hz).

Figure 11.

Attempts to form the free base of ammonium salt 32 resulted in decomposition, potentially through opening of the donor–acceptor cyclopropane (see inset).

We employed DFT calculations using the protocol of Hoye and co-workers⁴⁹ to determine the theoretical carbon-13 chemical shifts for all possible diastereomers of (+)-ocimicide A₂ at carbons 12, 13, 14, and 17 and N15 (Figure 12). Each of the 32 diastereomers was subjected to a conformational search using BOSS,⁵⁰ and the geometry of all conformers within 5 kcal/mol of the lowest-energy conformer was optimized. The carbon-13 shifts were calculated⁵¹ and Boltzmann-averaged. The accuracy of these calculations was benchmarked against 19-(E)-hunteracine (33)⁵² and the ammonium salt of 32 (Figure 12b).^51,53

Figure 12.

a. RMSE and absolute error (AE) of calculated carbon-13 chemical shifts for the reported diastereomer of ocimicide A₂ (20) and the average of all alternate diastereomers of (+)-ocimicide A₂ (20a) varied at positions 12, 13, 14, and 17. The numbering system employed corresponds to that used in the isolation reports. b. RMSE and AE of theoretical carbon-13 chemical shifts for 19-(E)-hunteracine (33) and Weinreb amide ammonium salt 32. c. Absolute difference between theoretical and reported carbon-13 chemical shifts for the reported structure of ocimicide A₂ (20).

A comparison of the reported and theoretical carbon-13 chemical shifts for (+)-ocimicide A₂ (20) revealed large inconsistencies, notably at the lactam–quinoline fusion [C9 and 20; absolute error (AE) = 7.3–13.1 ppm] and the cyclopropane (C10; AE = 4.2–9.8 ppm). These three positions were calculated for the salt 32 with high accuracy (AE = 0.3–1.8 ppm for C9 and 20; AE = 2.8 ppm for C10), suggesting a discrepancy between the actual structure and the assignment at this region of the molecule. Larger deviations were observed within the pyrrolidine ring of 20 (C13, 14, 22, and 23; AE = 8.3–20.9 ppm) and the diastereomers 20a (AE > 6.0 ppm for C14, 22, and 23), indicating that the spectroscopic differences were not exclusively of a stereochemical nature.

In the absence of both original spectra and 2D NMR data or access to the original isolates, we attempted to reisolate the material from various strains of Ocimum sanctum following the reported protocol but were unable to observe (by LC/MS) the alkaloids in the plant extracts. In light of this and with no clear alternative structure to computationally examine or synthesize, our efforts toward the ocimicide alkaloids came to an end. The observed reactivity and spectroscopic differences led us to conclude that a structural revision of these natural products is likely necessary.

DISCUSSION AND CONCLUSIONS

We end with some thoughts on the limitations and future of secondary metabolite structure determination and standards for data reporting, and we advocate for the systematic incorporation of carbon-13 DFT calculations into structure determination and synthesis.

MicroED

Our lomaiviticin studies underscore the significant impact that microED stands to have on secondary metabolite structure determination. In the absence of microED, it would have been difficult to advance the structure revision with confidence due to the ambiguity of multiple-bond C─H correlations. Additionally, the C₂ symmetry of (−)-lomaiviticins A and B makes it impossible to distinguish transannular ROESY and HMBC correlations from interactions of atoms within the same ring. Though introduced only in 2018,¹⁶ microED has already been employed in conjunction with state-of-the-art genomics methods to facilitate the identification of novel secondary metabolites and to correct additional structure misassignments.⁵⁴ The impact of this technique is certain to increase further as purpose-built microED instruments become widely available.

NMR Analysis

Recent advances in instrumentation allow NMR analysis to be carried out on <1 μmol quantities of sample.⁵⁵ However, the interpretation of these data can still be problematic, especially for highly oxidized metabolites such as the lomaiviticins. The original structure assignment relied heavily on HMBC correlations, and it can be difficult or impossible to distinguish ²J_C─H, ³J_C─H, or ⁴J_C─H couplings. A common misconception is that longer-range correlations display lower intensity than shorter-range correlations. In fact, the intensity of H─C couplings is largely unpredictable.⁵⁶ New methods for differentiating two- and three-bond HMBC correlations (e.g., separate echo and antiecho XLOC or SEA XLOC experiments)⁵⁷ as well as directly delineating carbon–carbon correlations (adequate sensitivity double-quantum spectroscopy, ADEQUATE)⁵⁸ have been developed. However, extended instrumentation time and optimization of the experiment are required to properly implement these methods.

The construction of a global infrastructure to support the dissemination of NMR data has been advocated for elsewhere,⁵⁹ and we echo this sentiment here. Most researchers have access to NMR processing software, and a database that provides FID files would be of immense utility. An extension of this database to the patent literature is warranted as well, though this is a more challenging objective. Automated methods for structure determination based on NMR data are being developed.⁶⁰

DFT NMR Calculations

We used the method of Hoye and co-workers⁴⁹ to calculate theoretical carbon-13 shifts in our ocimicide work. Though state of the art at the time, the approach still required a large amount of manual intervention. In our later lomaiviticin work, we employed Spartan,⁶¹ which has a fully automated method to calculate proton and carbon-13 NMR chemical shifts.¹⁸ The software uses iterative geometry optimizations and energy calculations to arrive at a small number of conformers for NMR calculations while minimizing computational time. As demonstrated in our lomaiviticin studies these methods provide accurate data for large structures. These computational methods are straightforward to execute, requiring the input of only a single candidate structure in Spartan. As such, we advocate for their systematic use in validating the assignments of novel secondary metabolites. We have also found them to be useful in synthetic studies to facilitate structure assignments of advanced intermediates when X-ray or microED data are not available.

Synthetic Chemistry in Conjunction with Functional Analysis

The natural product landscape has been transformed as a result of the rapid expansion of affordable next-generation sequencing technologies and data mining tools. These technologies have contributed to a surge in the discovery of novel BCGs and have assisted in the prediction of the metabolite chemical structure directly from sequencing data.⁶² Nonetheless, the process of isolating and functionally characterizing the encoded natural products remains a challenge. In the colibactin story, we demonstrated that the engagement of synthetic chemistry in the structure elucidation process can help to overcome some of the limitations inherent in isolation. By undertaking the targeted synthesis of isolated and putative metabolites, we obtained key insights into the structural assignment, stability, mechanism of action, and biosynthesis of colibactin. We propose that this multidisciplinary approach may enable the structural and functional analysis of other elusive metabolites.

Biographies

Mikaela DiBello was born in Mahopac, NY (1997), received her B.Sc. in chemistry from Rensselaer Polytechnic Institute in 2019, and is currently pursuing her Ph.D. at Yale University, where she has worked on the synthesis of novel pleuromutilin analogs and the total synthesis of diazofluorene natural products.

Alan R. Healy was born in Co. Clare, Ireland in 1988, completed his undergraduate studies in medicinal chemistry at Trinity College Dublin (TCD), and obtained an M.Sc. in biomedical science from the University of Edinburgh and a Ph.D. from St. Andrews University in 2014. Healy was a postdoctoral fellow with Seth B. Herzon and Jason M. Crawford at Yale University and is currently an assistant professor of chemistry at New York University Abu Dhabi (NYUAD). Healy’s research focuses on the development of novel methods to accelerate the discovery and study of dark matter metabolites.

Herman Nikolayevskiy was born in Tashkent, Uzbekistan in 1989, completed undergraduate studies in chemical engineering at The Cooper Union for the Advancement of Science and Art, obtained his Ph.D. in chemistry from Yale University 2017, and was an IRTA postdoctoral fellow at the NIH. He is currently an assistant professor and the program director for the MS in Chemistry Program at the University of San Francisco. Nikolayevskiy’s research focuses on the development of safer chemotherapies through triggered intramolecular deactivation and the development of covalent inhibitors against bacterial Sortase A.

Zhi Xu was born in Harbin, China in 1997. He obtained his undergraduate degree in chemical biology from Peking University under the supervision of Prof. Xiaoguang Lei. In 2019, he began his graduate studies in the laboratory of Prof. Seth Herzon at Yale University, where he works on total syntheses of diazofluorene and terpenoid natural products.

Seth B. Herzon was born in Philadelphia, PA in 1979, completed undergraduate studies at Temple University, obtained a Ph.D. from Harvard University, and was a postdoctoral fellow at the University of Illinois, Urbana–Champaign. He is currently the Milton Harris ’29 Ph.D. Professor of Chemistry and Professor of Pharmacology and Therapeutic Radiology at the Yale School of Medicine and a member of the Yale Comprehensive Cancer Center. Herzon’s research focuses on synthetic and translational studies of DNA damage and microbiome-derived secondary metabolites and the development of novel therapeutics targeting tumor-associated DNA repair defects.