Employing chemical synthesis to study the structure and function of colibactin, a “dark matter” metabolite.

Abstract

The classical discovery paradigm in the field of natural products follows the sequence of isolation, structure elucidation, chemical synthesis, and then elucidation of mechanism of action and structure–activity relationships. Although this discovery paradigm has proven successful in the past, researchers have amassed enough evidence to conclude that the vast majority of Nature’s secondary metabolites – biosynthetic “dark matter” – cannot be identified and studied by this approach. Many biosynthetic gene clusters (BGCs) are expressed at low levels, or not at all, and in some instances a molecule’s instability to fermentation or isolation prevents detection entirely. Here, we discuss an alternative approach to natural product identification that addresses these challenges by enlisting synthetic chemistry to prepare putative natural product fragments and structures as guided by biosynthetic insight. We demonstrate the utility of this approach through our structure elucidation of colibactin, an unisolable genotoxin produced by pathogenic E. coli in the human gut.

Graphical Abstract

This review recounts the chemical journey that culminated in the structural elucidation of the elusive bacterial metabolite, colibactin.

Introduction.

Secondary metabolites are the molecular tools produced and used by all organisms to sense and manipulate their environment and communicate with the organisms around them. Throughout history, humans have repurposed these molecules for applications in human health. Continued study of secondary metabolites will facilitate exploitation of these molecular resources for the improvement of the human condition, while also benefitting the fields of chemistry and biology broadly. Of these known natural products, bacterial secondary metabolites constitute the majority and possess a wide array of biological activities. However, despite significant progress in their isolation and characterization, it is believed that the more than 300,000 known bacterial secondary metabolites represent only a small fraction of molecules produced by these microorganisms.¹ The remaining uncharacterized secondary metabolites have been referred to as “dark matter” natural products.²

The classical paradigm of secondary metabolite discovery is predicated on the fulfillment of two criteria. The first criteria that must be met is production of sufficient amounts of material to allow for detection and isolation. Generally speaking, fulfillment of this criteria is rare, as the majority of BGCs are not expressed under laboratory conditions.³ Additionally, many bacteria are simply not amenable to laboratory culture, preventing large-scale production of secondary metabolites.⁴ The second criteria that must be met is stability towards fermentation, isolation, and purification. Fulfillment of this criteria can be near impossible to achieve in instances where solvent, water, or air exposure results in significant perturbation or degradation of the metabolites under study. As such, molecular instability poses a significant challenge to traditional isolation and characterization techniques. For metabolites where either of these two criteria are unfulfilled, the prevailing discovery paradigm generally fails.

Therefore, in the absence of major technological advances, the identification of unstable secondary metabolites will continue to rely on time- and labor- intensive processes such as isolating degradation products to infer the structures of their biosynthetic precursors, and then assembling evidence piece-by-piece until a reasonable structure can be proposed. An example of this type of approach is the study of the pathogenicity of the highly lethal Gram-negative Burkholderia mallei and Burkholderia pseudomallei, which was linked to a number of virulence factors including the product of the cryptic bur/mal gene cluster.^5-7 Identification of the mature toxin produced by this pathway proved to be troublesome, as isolated bur/mal metabolites including burkholderic acid (1)⁸ and sulfomalleicyprol (2)⁹ were found to be relatively non-toxic (Figure 1a). As such, Hertweck and co-workers suspected that these metabolites were degradation products of an unstable, mature bur/mal biosynthetic product. Through meticulous LC/MS molecular networking analysis, Hertweck and co-workers identified the suspected mass of the mature product of the bur/mal gene cluster. Careful optimization of purification conditions resulted in isolation of the mature toxin, bis(malleicyprol) (3), which was found to exist in equilibrium with malleicyprol (4) and iso-malleicyprol (5). Degradation studies found that bis(malleicyprol) (3) produces sulfomalleicyprol (2) and burkholderic acid (1) by conjugate addition of sulfite or cyclopropane ring-opening by hydroxide, respectively. The equilibrium mixture of 3, 4, and 5 was found to have a GI₅₀ of 1.14 μM against human umbilical vein endothelial cells, as compared to a GI₅₀ of 127 μM and 129 μM for 1 and 2, respectively. Analogous results with other human cell lines supported the theory that bis-malleicyprol (3) is the mature toxin of the bur/mal gene cluster.⁹ In this case, the instability of the mature toxin precluded initial isolation attempts. Ultimately, however, the mature toxin was not so unstable as to prevent isolation after optimization of the isolation procedure. It is important to note that more unstable metabolites might not be amenable to the same approach.

Figure 1.

a. Bis(malleicyprol) (3) degrades to burkholderic acid (1) and sulfomalleicyprol (2). Isolated 3 exists in equilibrium with monomers 4 and 5. b. TXA₂ (6) degrades to TXB₂ (7) with a half-life of 32 s at pH 7.4 and 37 °C. c. Pleurocybellaziridine (8) forms nucleophilic adducts such as 9–11 upon attempted isolation.

In cases where metabolite instability does completely prevent isolation, synthetic material can instead be utilized to validate the structure predictions. An early example of such a strategy can be found in Still and co-workers’ synthesis of the platelet aggregation factor TXA₂ (6, Figure 1b).^10,11 TXA₂ (6) possesses a half-life of 32 s at pH 7.4 and 37 °C, making isolation and characterization difficult. In 1975, Samuelsson and co-workers used the structure of the degradation product TXB₂ (7) to advance structure 6 for TXA₂, which contains a reactive and unstable oxetanyl acetal.¹² In 1985, Still and co-workers provided strong support for this prediction by synthesizing and studying 6, which was found to recapitulate the biological profile and half-life of the natural material.^10,11

Similarly, the Kawagishi lab’s studies of pleurocybellaziridine (8) capitalized on the utility of directly synthesizing the proposed metabolite.¹³ Although generally considered edible, consumption of the mushroom Pleurocybella porrigens was linked to fifty-five cases of poisoning in 2004, resulting in seventeen deaths by acute encephalopathy. In 2010, Kawagishi and co-workers investigated the molecular basis of toxicity of the P. porrigens fruiting body. They isolated several weakly cytotoxic metabolites (9–11) containing a β-hydroxyvaline fragment conjugated to various alcohols (Figure 1c). This led them to propose the existence of a common amino acid precursor containing a reactive aziridine moiety that renders the mature toxin pleurocybellaziridine (8) unstable and therefore unisolable. To probe for the existence of this hypothetical aziridine precursor, Kawagishi and co-workers developed a synthetic route to pleurocybellaziridine (8). These studies allowed the researchers to confirm the existence of this molecule in P. porrigens fruiting bodies.¹⁴ To validate their findings, it was shown that synthetic pleurocybellaziridine (8) does spontaneously decompose under conditions similar to those used in the isolation report. It was also determined that synthetic pleurocybellaziridine (8) is significantly more toxic to rat CG4-16 oligodendrocyte cells than other previously isolated metabolites. As such, the discovery and structure confirmation of both pleurocybellaziridine (8) and TXA₂ (6) are prominent examples of the utility of synthetic chemistry for the structure elucidation of unstable natural products.

Our lab has been heavily involved with the study and structure elucidation of colibactin, a “dark matter” genotoxic bacterial metabolite. Colibactin was discovered indirectly when strains of E. coli harboring the clb (or pks) BGC, which encodes its biosynthesis, were found to induce DNA double-strand breaks both in vitro¹⁵ and in vivo.¹⁶ Control experiments linked this phenotype to the final product of the clb pathway, colibactin, but researchers were unable to isolate and characterize this toxin.¹⁵ It was subsequently found that colibactin is produced by many pathogenic bacteria with the clb BGC being commonly associated with other bacterial virulence factors.¹⁷ When infected with clb⁺ bacteria, mouse models of colitis were observed to develop polyps and tumors, suggesting that colibactin could be an oncogenic driver within an inflamed gut.^18-20 Subsequent clinical studies supported this hypothesis by correlating the presence of clb⁺ strains of bacteria in the human gut microbiota with the frequency and severity of colorectal cancer (CRC).^18-21 More recently, DNA mutational signatures induced by colibactin have been characterized in an organoid model of the intestinal epithelium and these same mutational signatures were then detected in CRC patients.²² Recent studies have also detected clb metabolites in urine samples of patients with community-acquired urinary tract infections.^23,24 This growing body of data implicating colibactin in CRC development has sparked efforts to understand its structure and mechanism of action. In this review, we discuss how our approach employed chemical synthesis in conjunction with biosynthetic and mechanism of action studies to ultimately elucidate colibactin’s structure. The review is framed around key milestones in our own work along this pathway. For broader discussions the reader is directed toward several recent reviews.^25-30

Main text.

Synthetic chemistry resolves the structure of precolibactin A.

Initially, structure elucidation and isolation of colibactin through traditional means proved unsuccessful, presumably due to colibactin’s instability towards fermentation and isolation conditions. To circumvent this, many early studies involved culturing genetically-modified clb⁺ E. coli and analyzing the products in these cultures. Biosynthetic studies established that ClbP – a pathway-dedicated serine protease encoded by the clbP gene^31,32 – cleaves an N-myristoyl-d-Asn residue from biosynthetic intermediates known as precolibactins to reveal colibactin^33-36 (for a detailed discussion of the clb biosynthetic gene cluster and colibactin biosynthesis please see ref. 36). Researchers hypothesized that retention of the N-myristoyl-d-Asn residue by deactivation of ClbP might allow for isolation of precolibactins. It was further reasoned that characterization of the biosynthetic pathway in conjunction with structure elucidation of precolibactins would ultimately allow researchers to deduce colibactin’s structure.

Several precolibactins were successfully isolated and characterized from ΔclbP strains of E. coli. The lactam 12 was an advanced ΔclbP isolate that was fully characterized by standard spectroscopic methods (Figure 2).^37-39 Detection of a metabolite in ΔclbP extracts with m/z = 816.3780 in conjunction with the structure of 12 and biosynthetic analysis led Vizcaino and Crawford to predict the structures 13a and 13b, which were later named precolibactin A.³⁷ The presence of an N-terminal N-myristoyl-d-Asn residue was inferred based on the structure of previously isolated clb metabolites. The presence of heterocycles in the predicted structure was supported by MS analysis and feeding studies, which indicated the incorporation of two cysteine residues. However, these studies were unable to distinguish between the two possible isomers 13a and 13b, which differ in the connectivity of the thiazole and thiazoline rings.

Figure 2.

clb metabolite 12 was isolated and fully characterized by standard spectroscopic methods. The two isomeric structures 13a and 13b were predicted by the Crawford laboratory as the structure of precolibactin A.

Although precolibactin A was detectable in bacterial extracts, it readily decomposed upon attempted isolation.³⁷ As such, we sought to prepare synthetic standards of 13a and 13b and to determine if they corresponded to natural material by LC/MS co-injection with the bacterial extracts.⁴⁰ Our synthetic approach was to synthesize a fully linear precursor, followed by cyclodehydration to generate the lactam. Condensation of N-(tert-butoxycarbonyl) aminoacetonitrile and l-cysteine ethyl ester provided the thiazoline 14 (85%, Figure 3a). Amidation followed by treatment with Lawesson’s reagent generated the thioamide 15 (>99%, two steps). Condensation of the thioamide 15 with bromopyruvic acid formed the thiazole 16. Treatment with hydrochloric acid provided the amine 17 (70% over two steps). Silver coupling of the amine 17 with the β-ketothioester 18 (prepared in one step and 56% yield from N-(tert-butoxycarbonyl)-1-aminocyclopropane-1-carboxylate), followed by removal of the carbamate (hydrochloric acid), provided the amine 19. The isomeric amine 20 was prepared using a similar route (Figure 3b).

Figure 3.

a. Synthesis of the thiazoline–thiazole 19. b. Synthesis of the thiazole–thiazole 20. c. Synthesis of the N-myristoyl-d-Asn β-ketothioester 23. d. Fragment coupling to form the cyclization precursors 24 or 25.

The N-myristoyl-d-Asn fragment 23 was prepared as shown in Figure 3c. Amide coupling of commercial N-(tert-butoxycarbonyl)-d-asparagine (21) with (S)-hex-5-en-2-amine, followed by removal of the carbamate (hydrochloric acid), provided the alkene 22 (84%, two steps). N-Acylation with myristoyl chloride, followed by oxidative cleavage of the alkene (ruthenium trichloride, sodium periodate), generated a carboxylic acid (not shown) that was activated with carbonyldiimidazole (CDI) and then treated with 3-(tert-butylthio)-3-oxopropanoic acid to provide the desired N-myristoyl-d-Asn β-ketothioester 23 (74% over three steps). Finally, silver-mediated coupling of 23 with the isomeric amines 19 and 20 generated the linear precolibactin A precursors 24 and 25 (90% and 87%, respectively; Figure 3d).

With the linear precolibactin A precursors 24 and 25 in hand, we began to evaluate conditions to enact cyclization to their respective unsaturated lactams. Surprisingly, exposure of 25 to potassium carbonate in methanol at 0 °C did not result in the desired product 13b. Instead, the pyridone 26 arising from two-fold cyclodehydration was isolated (Figure 4a). Similar results were obtained when 25 was treated with ammonium carbonate in ethanol or aqueous sodium hydroxide. The presence of a pyridone moiety was supported by comparison with the ¹H NMR data of precolibactins B and C (27 and 28, respectively, Figure 4b), which had been isolated and characterized by standard spectroscopic methods.^41,42

Figure 4.

a. Two-fold cyclodehydration of the linear precursor 25 to the pyridone 26 occurs via the intermediacy of the unsaturated lactam 13b (predicted as precolibactin A). b. Structures of precolibactins B (27) and C (28). Both 27 and 28 were isolated and characterized by standard spectroscopic methods. C. Corrected structure of precolibactin A as 29.

We recognized that our discovery of a pyridone-forming cyclodehydration would enable us to synthesize and confirm the structures of precolibactins B (27) and C (28). Towards this end, we prepared the linear precursors corresponding to precolibactins B and C using routes analogous to that shown in Figure 3 (not shown). Subjecting these linear precursors to potassium carbonate in methanol generated synthetic precolibactin B (27) and precolibactin C (28), which were indistinguishable (by ¹H and ¹³C NMR analysis) from bacteria-derived material, thereby confirming their structural assignments.

Returning to the synthesis of precolibactin A, we found that the rate of undesired pyridone formation for 25 could be outcompeted by lactam formation by treatment with potassium carbonate in methyl sulfoxide at 24 °C. Thus, the conversion of 25 was allowed to proceed until a mixture of the lactam 13b and the pyridone 26 was obtained (ca. 7 h, see Figure 4a, right). Aqueous work-up followed by HPLC analysis established that 13b, presumed to be precolibactin A, was distinct from natural material. Similarly, the isomeric structure 13a deriving from 24 also did not co-elute with natural precolibactin A. As such, it was clear that both predicted structures of precolibactin A, 13a and 13b, were incorrect.

To resolve this, we carefully re-examined the initial prediction report.³⁷ We noted that the fermentation had been optimized by increasing the cysteine concentration five-fold. This raised the possibility that precolibactin A might be a derailment product wherein an enzyme-bound thioester intermediate is intercepted by exogenous cysteine, leading to a terminal N-acyl cysteine product. Additionally, given the ease with which the pyridone-forming two-fold cyclodehydration occurred (vide supra), we surmised that the pyridone residue present in precolibactins B (27) and C (28) might also be present in precolibactin A. Cognizant of these two structural possibilities, the structure 29 was proposed as precolibactin A (Figure 4c). Pyridone 29 is an isomer of the original predicted structure and was consistent with all available data, including the cysteine feeding studies previously conducted.

To test our proposed structural revision, we coupled synthetic precolibactin B (27) with l-cysteine (N-hydroxy succinimide and N-ethyl-N’-(3-dimethylaminopropyl)carbodiimide hydrochloride, 89%) to provide 29. We found that synthetic 29 was indistinguishable from bacterial-derived material by LC/HRMS-QTOF co-injection and tandem MS, thereby confirming our structure revision. NMR comparison with the natural material was not possible as all attempts to isolate precolibactin A (29) from bacterial extracts for characterization have proven unsuccessful,⁴⁰ a result we attribute to formation of mixed disulfides.

In summary, these studies established that two-fold cyclodehydration of linear precolibactins to form pyridones is a facile process. This discovery enabled the synthesis and structure confirmation of precolibactins B (27) and C (28). In addition, synthesis of the structures 13a and 13b allowed us to revise the structure of precolibactin A to 29. Importantly, our synthetic studies revealed that precolibactins A–C (27–29) all contain a substituted pyridone.

Synthetic chemistry reveals precolibactins as artifacts of clbP mutation.

Having established the presence of substituted pyridones in precolibactins A–C (27–29), our focus turned toward understanding if and how these pyridone-containing isolates contribute to colibactin’s genotoxicity. Others had previously speculated that the pyridone moiety plays a role in genotoxicity.⁴¹ At the time that precolibactin A was thought to contain an unsaturated lactam (13a or 13b), a mechanism of action was proposed^37,39 that initiates with ClbP deacylation (which was known to be required for genotoxicity)^15,33,34 followed by DNA alkylation via cyclopropane ring-opening (see 30, Figure 5a). Possibly, the unsaturated iminium ion 31, formed by cyclodehydration of 30, might generate a more potent conjugated cyclopropane electrophile. It was also suggested that multiple alkylations could generate a DNA interstrand cross-link.³⁷

Figure 5.

a. When it was believed that precolibactin A contained an unsaturated lactam (as in 13a or 13b), it was suggested³⁷ that the mechanism of genotoxicity involved ClbP deacylation and nucleotide addition to 30 or 31. To a first approximation, nucleophilic addition to the cyclopropane in precolibactins A–C (27–29) seemed less plausible. b. We postulated that the pyridones 27–29 formed due to the persistence of the N-myristoyl-d-Asn residue in clbP mutant strains. We proposed that in wild-type strains, ClbP deacylation may be competitive, diverting the cyclization pathway and resulting in formation of the unsaturated imine 31.

After completing our initial synthetic studies, we knew that precolibactins A–C (27–29) all contain a pyridone residue. To a first approximation, we expected that nucleophilic addition to the cyclopropane in these pyridone-containing structures (as opposed to the cyclopropanes in unsaturated imine-containing structures such as 30 or 31) would not be facile as nucleophilic attack and concomitant ring opening of the cyclopropane would result in the unfavorable dearomatization of the pyridone ring. Thus, we hypothesized either that the mechanism of action hypothesis was incorrect or that an important detail had been overlooked.

Contemporaneously, our collaborators in the Crawford laboratory had accumulated genetic evidence suggesting that linear biosynthetic products such as 32 (analogous to those chemically synthesized earlier) are off-loaded from the biosynthetic assembly line and that the ensuing cyclodehydration reactions are not enzyme-catalyzed (Figure 5b).⁴³ Because all precolibactins had been detected in and/or isolated from cultures of clbP mutants, we surmised that the persistence of the N-myristoyl-d-Asn residue was diverting these intermediates toward pyridone formation (see 32→27–29). In wild-type strains, ClbP deacylation possibly outcompetes cyclization, leading instead to the primary amine 34. This amine would be expected to be substantially more nucleophilic than the amide in 32 and 33. Accordingly, an alternative mode of cyclization (34→35→31) might divert the pathway toward the iminium ion 31. If this were correct, the original mechanism of action hypothesis could still be valid, with pyridone-containing precolibactins simply being derailment products deriving from mutation of clbP.

At this time, unsaturated imines such as 31 had not yet been detected in bacterial extracts. Consequently, we set out to synthesize them, as well as the corresponding deacylated pyridone, in order to probe this hypothesis.⁴⁴ Accordingly, we synthesized the N-(tert-butoxycarbonyl)-protected linear precursor 36 by a slight modification to our earlier synthetic route (Figure 6). We found that compound 36 could be selectively cyclized to either the pyridone 37 (78%) or the vinylogous amide 39 (87%) upon treatment with potassium carbonate in methanol or concentration from dilute acid, respectively. Amidation of 37 (propylphosphonic acid anhydride (T₃P), N,N-dimethylethylenediamine (DMEDA) and carbamate cleavage (trifluoroacetic acid) provided the target pyridone 38 (40%, two steps). Likewise, amidation of 39 (T₃P, DMEDA) generated the amide 40 (95%). Carbamate cleavage (trifluoroacetic acid) followed by neutralization with aqueous sodium bicarbonate solution yielded the unsaturated imine 41 (62%).

Figure 6.

Synthesis of the pyridone 38 and the unsaturated imine 41.

We evaluated the DNA-damaging ability of compounds 38 and 41 under cell-free conditions. Linearized plasmid pBR322 DNA was incubated with either compound for 15 h at 37 °C, and the resulting DNA was analyzed by denaturing gel electrophoresis. While the pyridone 38 did not generate detectable levels of DNA damage at concentrations up to 0.5 mM, the unsaturated imine 41 extensively damaged DNA at concentrations as low as 100 nM. We conducted a DNA-binding assay to verify that these observations do not arise from differences in DNA binding affinity. Treatment of calf thymus DNA (ctDNA) with the pyridone 38 resulted in an increase in melting temperature (T_m) of 2.8 °C at a ratio of 1:2 38:base pairs. On the other hand, incubation of ctDNA with the imine 41 resulted in a time-dependent change in T_m. Shorter incubation times (5 min) resulted in a T_m increase of 0.95 °C at a ratio of 1:2 41:base pairs, whereas longer incubation times (3 h) showed an overall decrease in T_m. We interpreted this as an initial binding of imine 41 to DNA, followed by alkylation and slow degradation of the duplex. Thus, 38 and 41 bind DNA with comparable affinity.

A series of compounds were prepared to gain further insight into the molecular basis of DNA alkylation (Figure 7). The gem-dimethyl derivative 42 was prepared to probe for involvement of the cyclopropane in DNA alkylation. As anticipated, 42 exhibited no detectable DNA alkylation activity at concentrations as high as 0.5 mM, confirming that the conjugated cyclopropane is essential. Furthermore, we reasoned that if the conjugated cyclopropane 41 is capable of alkylating DNA, synthesis of a dimer, such as 43, would enable formation of an intrastrand crosslink (ICL). We synthesized 43 and observed ICL formation after incubating 43 (10 μM) with linearized plasmid DNA for 3 h at 37 °C. Additionally, we observed a positive correlation between the amount of DNA damage and the degree of positive charge at the amide extension, as in 44a–e, with tertiary amines damaging DNA most extensively. Finally, we investigated the extent to which the cyclic imine impacts the electrophilicity of the cyclopropane. In order to prevent pyridone formation, the N-methyl derivative 45 was synthesized. As a control, the N-methyl unsaturated imine 46 was also prepared. Interestingly, the lactam 45 showed weak DNA alkylation activity at 500 μM, but the unsaturated imine 46 proved much more potent, alkylating DNA at concentrations as low as 1 μM.

Figure 7.

Synthetic analogs designed to probe the molecular mechanism of DNA damage. The gem-dimethyl derivative 42 is not genotoxic, indicating the cyclopropane is essential for activity. The dimer 43 generates DNA ICLs, supporting DNA damage by DNA alkylation at the cyclopropane. Analogs (44a–e) with increased positive charge density are more potent DNA alkylators. The unsaturated imine 46 is a more potent DNA electrophile than the lactam 45. The N-methyl substituent (blue in 45 and 46) was introduced to avoid any complications arising from pyridone formation.

Together, these studies suggested that the genotoxicity of colibactin arises from DNA alkylation by a structurally unprecedented cyclopropane warhead (for reviews of alkylation of DNA by other cyclopropane-containing natural products, see refs. ^45-47). These studies also established that pyridone-containing clb metabolites do not alkylate DNA. Moreover, these studies support the notion that pyridone-containing precolibactins, such as precolibactin A–C (27–29), are not biosynthetic precursors to colibactin but instead derive from mutation of clbP. It is noteworthy that at the time these studies were conducted, natural unsaturated imines had not been characterized in any form from clb cultures. Had these synthetic studies not been conducted, it is possible that the full effects of ClbP deactivation on metabolite structure may not have come to light.

Synthesis of precolibactin 886 reveals it may be an artifact of analysis and purification.

Precolibactin 886 (47) was isolated by Qian and co-workers from a clbP/clbQ double mutant in low yield (2.8 mg of 47 were obtained from a 1000-L fermentation; Figure 8).⁴⁸ Precolibactin 886 (47) contains a unique 15-atom macrocycle and is one of the most biosynthetically advanced isolated precolibactins, with only three enzymes (ClbL, ClbO, and ClbQ) unaccounted for in its biosynthesis. Considering the structural differences between precolibactin 886 (47) and precolibactins A–C (27–29), as well as the report that precolibactin 886 (47) is modestly cytotoxic (IC₅₀ = 22.3 and 34.0 μM against HCT-116 and HeLa cell lines, respectively),⁴⁸ we sought to study the chemistry of this isolate via total synthesis.⁴⁹ As previous synthetic studies of precolibactins provided insights into colibactin’s biosynthesis in unique and unexpected ways, we envisioned that our synthetic work might illuminate new aspects of clb metabolite reactivity.

Figure 8.

Total synthesis of precolibactin 886 (47) and its degradation to precolibactin B (27).

Synthesis of precolibactin 886 (47) began with bifurcation of the thiazole 48 (accessible in one step and 74% yield from a commercial material) via reduction with di-iso-butylaluminum hydride to afford the aldehyde 49, or carbamate cleavage (hydrochloric acid) and transimination with benzophenone imine to afford the imine 50. A silver-mediated coupling then provided the 1,2-aminoalcohol 51 as an inconsequential mixture of diastereomers. Wong diazo transfer⁵⁰ provided the α-azidoalcohol 52 (56% over three steps), which was deprotected and coupled with the β-ketothioester 18 to generate 53 (85%, two steps). Saponification (lithium hydroxide) and acidic deprotection (hydrochloric acid) formed the aminocyclopropane 54 (87%, two steps). Coupling with the N-myristoyl-d-Asn β-ketothioester 23 followed by Staudinger reduction of the azide (trimethylphosphine) generated the α-aminoalcohol 55 (78%, two steps). Treatment of 55 with 2-iodoxybenzoic acid in methyl sulfoxide-d₆ at 23 °C resulted in consumption of 55 within 30 min and formation of the α-ketoimine 56.

With the linear ketoimine 56 in hand, we began to explore conditions to achieve macrocycle formation. Unfortunately, cyclization was not observed (by ¹H NMR spectroscopy) under a broad range of thermal, acidic, or basic conditions. These experiments were constrained by the instability of 56. For example, warming 56 under acidic conditions promoted decomposition and hydrolysis of the imine to the diketone 57. In light of our inability to induce cyclization, as well as the strong biosynthetic evidence supporting 56 as the precursor to precolibactin 886 (47),⁴³ we began to suspect that cyclization was occurring upon purification and analysis. To test this, we subjected the unpurified α-ketoimine 56 to semi-preparative HPLC. We found that under these conditions precolibactin 886 (47) could be isolated in 3% yield. Attempts to induce macrocyclization by exposing the α-ketoimine 56 to the HPLC solvent before purification under anaerobic conditions were unsuccessful.

These surprising results raised questions as to the origin and significance of precolibactin 886 (47). Considering that all cultures of clb⁺ E. coli have, to the best of our knowledge, been analyzed and purified by reverse-phase HPLC, it is possible that the linear precursor to precolibactin 886 (47) only cyclizes upon HPLC analysis. On the other hand, this would necessitate that the α-ketoimine 56 form and persist for a finite time in the aqueous enviornment of the fermentation media. As such, further studies are required to rigorously determine the origin of precolibactin 886 (47).

Discovery of an unexpected carbon–carbon bond cleavage pathway and implications for clb biosynthesis.

The low yield of precolibactin 886 (47) obtained after purification led us to question the fate of the linear precursor 56 on the HPLC column. We had obtained anecdotal evidence that 56 readily decomposes to smaller fragments, although these had not yet been characterized. Because synthetic intermediates lacking the α-ketoimine or α-dicarbonyl functional groups were stable toward normal laboratory manipulations, we suspected that these functional groups might be the source of instability. As such, we prepared the truncated systems 59 and 60 shown in Figure 9a.⁴⁹ This work led to the unexpected discovery that the C36–C37 bond in 56 and 57 is unstable toward nucleophilic cleavage under mild conditions, a finding that had several important implications for the field (vide infra).

Figure 9.

a. Oxidation of the aminoalcohol 58 provides the α-ketoimine 59, which hydrolyzes to the α-diketone 60. b. Exposure of the N-(tert-butoxycarbonyl)-aminoalcohol 62 to substoichiometric amounts of oxidant provides the α-ketoamine 63, which was found to spontaneously transform to the N-acyl hemiaminal 64 under air. c. Treatment of the α-ketoimine 59 with various nucleophiles results in cleavage of the bond joining the ketone and imine functional groups (C36–C37 in precolibactin 886 (47), see Figure 8).

The aminoalcohol 58 was prepared by a route analogous to that shown in Figure 8. Two-fold oxidation with excess Swern reagent provided the α-ketoimine 59 (nominally 93% yield), as determined by ¹H NMR analysis. The α-ketoimine readily hydrolyzed to the α-dicarbonyl 60 upon exposure to aqueous conditions. Extensive efforts to achieve a single oxidation to form 61 were unsuccessful. To constrain the reactivity of the ketoamine, we synthesized the N-(tert-butoxycarbonyl)-α-aminoalcohol 62 (Figure 9b). In the presence of excess oxidant, the expected N-(tert-butoxycarbonyl) ketoamine 63 was not observed; instead, the N-acyl hemiaminal 64 formed exclusively. However, treatment with 0.9 equiv of the Dess–Martin periodinane provided the desired N-(tert-butoxycarbonyl)-α-aminoketone 63. This species underwent spontaneous oxidation to the N-acyl hemiaminal 64 by air. Based on biosynthetic analysis it is expected that the α-aminoketone residue is formed initially in colibactin biosynthesis. An interesting implication of our findings was that oxidation to the α-ketoimine found in precolibactin 886 (47), the α-dicarbonyl in colibactin (vide infra), and other advanced intermediates, which had previously been attributed to ClbK, may instead occur spontaneously.⁴⁸

Next, we evaluated the stability of the α-ketoimine 59 and α-diketone 60 in the presence of various nucleophiles (Figure 9c). Treatment of the α-ketoimine 59 with sodium bicarbonate in methanol resulted in hydrolytic cleavage to form the ester 64 (42%) and the carboxamide 65 (42%), as well as other unidentified decomposition products. We found that solutions of the ketoimine in methanol-d₄ alone slowly underwent carbon–carbon bond cleavage (t_1/2 ~ 12 h at 24 °C). The carboxylic acid 66 was formed in 51% yield when the ketoimine 59 was exposed to sodium bicarbonate in aqueous tetrahydrofuran. Similarly, the amide 67 was formed in 36% yield upon treatment with excess pyrrolidine in tetrahydrofuran. In both of these cases, the fate of the right-hand portion of the molecule was not determined. We repeated these experiments with the model diketone 60 and found that it transformed in a strictly analogous manner, providing carboxylic acids, esters, or amides depending upon the nucleophile (data not shown).

Re-examination of the LC/MS and tandem MS spectra of the fully-elaborated α-ketoimine 56 revealed the formation of products analogous to those observed in the model studies (Figure 10). The linear carboxylic acid 68, arising from nucleophilic cleavage of the C36–C37 bond, was observed to undergo two-fold cyclodehydration to precolibactin B (27). Likewise, the addition of l-cysteine provided 69, which cyclized to precolibactin A (29). These findings suggested that precolibactins A (29) and B (27) might derive in part from C36–C37 degradation of more advanced biosynthetic intermediates. More importantly, however, these data provided an explanation for the inability to isolate fully-functionalized clb products. The remaining uncharacterized enzymes, ClbL and ClbO, are thought to elaborate the C-terminus of 56. C36–C37 bond cleavage would eliminate these functionalized fragments.

Figure 10.

Reexamination of the LC/MS data revealed that the α-ketoimine 56 degrades to precolibactins A (29) and C (27) in the presence of l-cysteine and water, respectively.

In summary, the study of the α-ketoimine 59 and the α-diketone 60, as well as precolibactin 886 (47) itself, provided further insight into the structures and reactivities of clb intermediates. We discovered that the C36–C37 linkage in advanced biosynthetic intermediates is unstable towards nucleophilic attack. Degradation of this bond leads to other known clb isolates and biosynthetic precursors, such as precolibactins A (29) and B (27). Our findings suggested that this reactivity renders advanced clb products unstable, thereby providing a simple molecular basis for the difficulties encountered in isolation efforts. The discovery that the α-ketoamine 55 oxidizes spontaneously to the α-ketoimine 56 suggests that the α-ketoamine formed in the biosynthetic pathway may be oxidized by air rather than enzymatically.

Synthesis validates the predicted structure of colibactin.

In order to circumvent colibactin’s instability, our lab devised a novel approach that exploits the relative stability of colibactin–DNA cross-links to facilitate structure elucidation.^36,51 This approach was founded on a report by Oswald and Nougayrède, which established that the genotoxicity of clb⁺ E. coli derives from the induction of DNA ICLs and that the resulting ICLs are relatively stable.⁵² Building on these critical findings, we sought to structurally characterize these ICLs as a means to deduce the structure of colibactin.

Towards this end, linearized pUC19 DNA was exposed to wild-type clb⁺ E. coli, the DNA was digested, and the resulting DNA adducts were analyzed by tandem MS. Using this tandem MS data, our lab was able to identify the mono-adenine adduct 70 (Figure 11).^51,53 The structure of this adduct was established by isotope labeling using l-[U-¹³C]-Cys and l-[U-¹³C]-Met, and extensive tandem MS analysis.

Figure 11.

Colibactin–DNA adducts detected by tandem MS from incubation of linearized pUC19 DNA with wild-type clb⁺ E. coli.

Subsequently, we detected a higher molecular weight ion (ultimately assigned as structure 71) and conducted extensive labelling studies to gain insights into its structure.³⁶ By comparing the resulting mass difference between the labeled auxotroph colibactin–DNA adduct and the unlabeled wild-type colibactin–DNA adduct, the composition of colibactin could be inferred. The presence of two thiazoles and two cyclopropane residues was established by the addition of l-[U-¹³C]-Cys and l-[U-¹³C]-Met to cultures of cysteine or methionine auxotrophs, resulting in mass shifts of six and eight units respectively. The addition of l-[U-¹³C]-Ser or l-[U-¹³C, ¹⁵N]-Ser to serine auxotrophs revealed that two distinct serine-derived fragments are incorporated into colibactin’s structure. The addition of d-[U-¹³C]-glucose to wild-type clb⁺ E. coli grown in media lacking amino acids suggested the presence of 37 carbon atoms in the colibactin residue. Similarly, the addition of [¹⁵N]-ammonium chloride indicated the presence of 8 nitrogen atoms, and a molecular formula of C₃₇H₃₈N₈O₉S₂ was therefore established for colibactin.

Careful consideration of the tandem MS data and biosynthetic knowledge suggested 71 as the predicted structure of bis(adenine) colibactin adduct. The monohydrate 72 was also observed. By inference, then, the structure of colibactin was predicted as 75. Mining the LC/MS data for doubly charged ions was essential to detect the bis(adenine) adduct 71 and the monohydrate 72, as the singly-charged ions corresponding to 71 and 72 are several orders of magnitude less abundant. Further analysis of the HRMS signals led to the identification of additional mono-adenine adducts such as the methyl-amino ketone 73, which results from premature off-loading from the biosynthetic pathway, as well as the right-hand fragment 74, which results from hydrolytic cleavage of the C36–C37 bond.

To confirm our proposed structure, we sought to obtain MS data for DNA cross-links derived from synthetic colibactin, and to compare these to those derived from the bacteria. Elaborating on our previous synthetic studies,^40,44,49 we envisioned a synthetic approach employing the colibactin precursor 76 as the key synthetic intermediate, which was prepared by a double acylation of the α-silyloxyketone 77 with two equivalents of the β-ketothioester 78 (Figure 12a). Synthesis of the requisite masked diketone 77 proceeded via a Riley-like oxidation of commercial thiazole ester 79, providing the aldehyde 80 (46%, Figure 12b). Reduction of the aldehyde (sodium borohydride) and saponification of the ester (lithium hydroxide) provided the carboxylic acid 81. Activation with 1,1′-carbonyldiimidazole (CDI) followed by addition of sodium methyl nitronate provided the α-nitroketone 82 (41%, three steps). Hydrogenolysis of the nitro group (dihydrogen, palladium on carbon) followed by acylation of the resulting amine (di-tert-butyldicarbonate) and oxidation of the primary alcohol (2-iodoxybenzoic acid, IBX) provided the aldehyde 83 (46% over three steps). Silyl cyanohydrin formation (tert-butyldimethylsilyl cyanide, imidazole) provided the umpolung coupling partner 84 (80%), which was coupled with the aldehyde 49 via a benzoin addition to provide the α-silyloxy ketone 85 (53%). Selective removal of the carbamate protecting groups (hydrochloric acid in dioxane) provided the bis(ammonium) salt 76 (99%). The β-ketothioester 77 was prepared by a silver-mediated coupling of the β-ketothioester 86 and the cyclopropyl ester 87 (Figure 12c). Cyclodehydration of the resulting product (not shown) then formed the vinylogous amide 88 (82%, two steps). Treatment of 88 with the enolate of tert-butyl thioacetate then provided the desired β-ketothioester 77 (79%).

Figure 12.

a. Our convergent synthetic strategy to access the masked diketone 75 via coupling of the diamine 76 and the β-ketothioester 77. b. Synthesis of the masked diketone 76. c. Synthesis of the β-ketothioester 77.

Silver-mediated coupling of the bis(ammonium) salt 76 with the β-ketothioester 77, followed by silylation in situ, formed the colibactin precursor 76 (17% yield by ¹H NMR analysis, 8.5% following semi-preparative HPLC purification, Figure 13). Global deprotection with hydrochloric acid in ethanol followed by aerobic oxidation yielded the diketone 89. Finally, treatment of this compound with rigorously deoxygenated citrate buffer (pH = 5) provided synthetic colibactin 75.

Figure 13.

Silver trifluoroacetate-mediated coupling of the masked diketone 77 and the β-ketothioester 78, followed by acidic deprotection and cyclization in aqueous buffer yields synthetic colibactin (75).

With synthetic colibactin 75 in hand, we characterized the ICLs derived from treatment of linearized pUC19 DNA with synthetic material to ICLs derived from natural material formed in clb⁺ E. coli. Synthetic colibactin (75) was found to generate ICLs as expected. After digestion and analysis by tandem MS, we found that ICLs derived from synthetic colibactin were indistinguishable from those derived from the bacteria, confirming our structural prediction. The same structure was advanced independently^54,55 by also mining HRMS data for doubly-charged (z = 2) ions. Currently, the exact nature of the C36–C37 linkage in colibactin at the time of ICL formation is an unresolved issue. As discussed earlier, the C36-C37 linkage is expected to exist at the α-aminoketone oxidation state following off-loading from the biosynthetic pathway. It has not been determined whether oxidation occurs before or after ICL formation.

Reflections on the journey to colibactin’s structure, conclusion, and outlook.

As discussed in the introduction, traditional natural product isolation relies on analytical purification to obtain homogeneous samples of secondary metabolites for structure elucidation. While exceptions exist, this approach has dominated the field for several decades. Colibactin constitutes an example of a secondary metabolite that is beyond the scope of this approach.

Beginning in 2015, we pursued the synthesis of clb metabolites fully aware that the complete structure of colibactin was unknown. Our work provided several essential insights into the clb pathway that ultimately facilitated the elucidation of colibactin’s structure. We developed a synthetic approach to access precolibactins A–C, establishing via synthesis that the predicted structure of precolibactin A (13a/b) was incorrect and correctly reassigning it as 29. The facile two-fold cyclization to produce pyridones prompted further study that led to the realization that deactivation of ClbP was derailing the biosynthetic pathway, leading to the production of pyridone-containing metabolites such as 27–29. This discovery led us to probe the DNA alkylating activity of these pyridones, which we found were not potent DNA alkylating agents. We then used synthetic chemistry to study unsaturated imines such as 41, which had previously been proposed as DNA-reactive agents. Our studies confirmed that these imines are in fact potent alkylation agents. Notably, imine-containing clb metabolites had not been identified in bacterial cultures at the time we conducted these studies, demonstrating the utility of synthetic chemistry as a means to hasten the discovery process.

Next, we used synthesis to perform mechanism of action studies by constructing key control compounds such as dimer 43 (which formed ICLs) and gem-dimethyl derivative 42 (which was inactive). The results of these experiments strongly suggested that the molecular basis for genotoxicity involves DNA alkylation by nucleotide addition into an α,β-unsaturated imine-conjugated cyclopropane. We then used synthesis to elucidate the structure of precolibactin 886 (47). Our studies raised the intriguing possibility that this “metabolite” is actually an artifact of the analytical and purification processes. Furthermore, our studies led us to identify an unexpected decomposition pathway involving carbon–carbon bond cleavage that suggested a simple explanation for the difficulties associated with isolation of advanced clb metabolites, including colibactin itself. Finally, we were able to amalgamate this chemical data, along with the wealth of biosynthetic knowledge of the clb pathway, to structurally characterize colibactin-derived ICLs and confirm our assignment via synthesis. Thus, our use of synthetic chemistry as a key component of the discovery process facilitated biosynthetic approaches towards understanding colibactin’s structure, biosynthesis, and its molecular basis for genotoxicity. Importantly, many of the insights we obtained would not have been achieved without engaging synthetic chemistry early in, and during, the discovery process.

We believe that the marriage of synthetic chemistry and biosynthetic analysis was crucial to the structure elucidation of colibactin. Our application of synthesis early on in the discovery process provided unique insights into several aspects of clb metabolite biosynthesis and reactivity. We believe that the success of this multidisciplinary approach is a testament to the strength of collaborative research. Specifically, one must acknowledge that the wealth of clb biosynthetic data generated over more than a decade by laboratories worldwide was essential to the structure elucidation of colibactin. Considering this, and recognizing that this level of biosynthetic data is generally not available for other “dark matter” metabolites, one is left to ponder the question: are generalizable approaches to the study of “dark matter” natural products achievable? In the absence of abundant biosynthetic data, can a minimum data set that allows for characterization of a particular “dark matter” metabolite structure and function be defined?

In our case, the structure elucidation of colibactin was ultimately achieved by harnessing the unique reactivity of the metabolite and its biosynthetic precursors. Although we acknowledge that our work is far from a general approach towards the structure elucidation of “dark matter” metabolites, it is our hope that our marriage of synthesis and biosynthetic analysis serves to inspire efforts to generalize and simplify the study of “dark matter” metabolites.