Enzymatic C_β-H functionalization of l-Arg and l-Leu in nonribosomally-derived peptidyl natural products: a tale of two oxidoreductases

Abstract

Muraymycins are peptidyl nucleoside antibiotics that contain two C_β-modified amino acids, (2S,3S)-capreomycidine and (2S,3S)-β-OH-Leu. The former is also a component of chymostatins, which are aldehyde-containing peptidic protease inhibitors that—like muraymycin—are derived from nonribosomal peptide synthetases (NRPSs). Using feeding experiments and in vitro characterization of twelve recombinant proteins, the biosynthetic mechanism for both nonproteinogenic amino acids is now defined. The formation of (2S,3S)-capreomycidine is shown to involve an FAD-dependent dehydrogenase:cyclase that requires an NRPS-bound pathway intermediate as a substrate. This cryptic dehydrogenation strategy is both temporally and mechanistically distinct in comparison to the biosynthesis of other capreomycidine diastereomers, which has previously been shown to proceed by C_β-hydroxylation of free l-Arg catalyzed by a member of the non-heme Fe²⁺- and α-ketoglutarate (αKG)-dependent dioxygenase family and (eventually) a dehydration-mediated cyclization process catalyzed by a distinct enzyme(s). Contrary to our initial expectation, the sole non-heme Fe²⁺- and αKG-dependent dioxygenase candidate Mur15 encoded within the muraymycin gene cluster is instead demonstrated to catalyze specific C_β hydroxylation of the Leu residue to generate (2S,3S)-β-OH-Leu that is found in most muraymycin congeners. Importantly, and in contrast to known l-Arg-C_β-hydroxylases, the Mur15-catalyzed reaction occurs after the NRPS-mediated assembly of the peptide scaffold. This late-stage functionalization affords the opportunity to exploit Mur15 as a biocatalyst, proof of concept of which is provided.

Graphical Abstract

INTRODUCTION

C_β-H functionalization can dramatically alter the physicochemical and biological properties of peptide-containing natural products and is often the difference between having an active molecule within a biological milieu that can be exploited for a desired purpose versus one that is inconsequential.^1–3 As would be expected, significant efforts have been invested in discovering and developing abiotic and enzyme catalysts to achieve C_β-H functionalization in a selective and efficient manner. Muraymycins, a group of antibiotic natural products isolated from Streptomyces sp. NRRL 30471 and mutant strains thereof, highlight the importance of C_β-H functionalization for antibacterial activity.^4–7 They are structurally categorized as peptidyl-nucleoside hybrids with the pseudopeptide component originating from three amino acid building blocks and containing a ureido group that reverses the peptide orientation to give, in essence, two C-termini, hence the use of “pseudo” terminology (Figure 1a). In addition to the unusual ureido group, two of the three amino acids are modified at C_β to give (2S,3S)-capreomycidine (1) and (2S,3S)-β-OH-Leu (2). The former, referred to as epicapreomycidine in past reports,⁵ is found in all congeners of muraymycin as well as the aldehyde-containing protease inhibitors chymostatins (Figure 1b), which like the muraymycins contain a ureido group at N_α of (2S,3S)-1 that changes the peptide orientation.⁸ The latter hydroxylated amino acid 2 is differentially modified to generate a series of muraymycin congeners labeled A-D, which were identified by fermentation of N-methyl-N′-nitro-N-nitrosoguanidine-derived mutant strains.⁴ The D series, represented by muraymycin D1 (3), muraymycin D2 (4), and muraymycin D3 (5), is the primary product of mutant strain Streptomyces sp. NRRL 30475 and contain an unmodified l-Leu; while the C series—the primary product from a different mutant strain called Streptomyces sp. NRRL 30477—contain (2S,3S)-2. The β-OH group can be differentially acylated leading to the A or B series. Antibacterial screens along with inhibitory tests examining peptidoglycan formation have revealed the general trend in potency of A ~ B > C ~ D series.^4,5 Consequently, C_β-H functionalization in this case not only leads to more than twenty unique muraymycin congeners but also enables the production of the most potent antibiotics of the family.

Figure 1.

Structures of representative 1-containing and related natural products. (a) The structure of the (2S,3S)-1-containing muraymycins and differences between the series. (b) The structure of representative chymostatins that contain (2S,3S)-1 and representative analogues that contain unmodified l-Arg instead of (2S,3S)-1. (c) The mostly established (viomycin) and proposed (faulknamycin) biosynthetic pathways leading to the respective diastereomers, (2S,3R)-1 and (2R,3S)-1. Whether amide bond formation with the upstream amino acid occurs prior to (R = H) or after (R = Val) conjugate addition during faulknamycin biosynthesis is unknown.

Capreomycidine [(2S,3R)-1] is a characteristic structural component of the antimycobacterial peptides viomycin⁹ and capreomycin¹⁰ and is known to originate from l-Arg through two catalytic steps (Figure 1c). A non-heme Fe²⁺- and α-ketoglutarate (αKG)-dependent dioxygenase VioC first catalyzes stereoselective C_β hydroxylation to form (2S,3S)-β-OH-Arg, and a pyridoxal-5’-phosphate-dependent enzyme VioD then catalyzes dehydration followed by intramolecular conjugate addition.^11–13 Once formed, (2S,3R)-1 is used as a substrate by a nonribosomal peptide synthetase (NRPS)—demonstrated with the NRPS CmnG involved in capreomycin biosynthesis¹⁴—for incorporation into the peptide. A new peptide natural product called faulknamycin was reported in 2020 that contains the capreomycidine enantiomer, (2R,3S)-1 (Figure 1c).¹⁵ This d-amino acid also originates from l-Arg, and like the biosynthesis of (2S,3R)-1, a non-heme Fe^2+-and αKG-dependent dioxygenase FauH has been proposed to first generate (2S,3S)-β-OH-Arg. However, the gene cluster for faulknamycin does not encode for a VioD homolog and, based on successful heterologous production of faulknamycin in Streptomyces lividans TK66 whose genome likewise does not encode for a homolog, VioD activity was speculated to be unnecessary for (2R,3S)-1 biosynthesis. Consequently, it was hypothesized that an adenylation (A) domain of the NRPS FauG loads (2S,3S)-β-OH-Arg to a thiolation (T) domain, and an upstream epimerase (E) domain—potentially in concert with a condensation (C) domain—catalyzes α-epimerization and C_β-N_η bond formation via dehydration and conjugate addition. Very recently, however, a second faulknamycin producer was characterized, and a VioD homolog (XNR1347) encoded outside the biosynthetic gene cluster was shown to be important for biosynthesis.¹⁶ Thus, the identity of the cyclization catalyst that generates (2R,3S)-1, and whether this occurs prior to or during NRPS-mediated peptide assembly, remains uncertain. Nonetheless, the current 1 biosynthetic paradigm irrespective of the stereochemical outcome involves C_β hydroxylation of l-Arg in a reaction catalyzed by a non-heme Fe²⁺- and αKG-dependent dioxygenase, followed by dehydration and conjugate addition catalyzed by a distinct enzyme to generate the capreomycidine ring.

Although hydroxylation is cryptic during the biosynthesis of 1, C_β-OH amino acids are common structural components in many peptidyl natural products.^17–19 Notably, in addition to the muraymycins, 2 is found in cyclopeptide alkaloids;²⁰ many cyclic depsipeptides including lysobactin,²¹ papuamides,²² skyllamycin,^23,24 and azinothricins;²⁵ the neurotrophic agent lactacystin;²⁶ among many others. The biosynthesis of 2 could potentially involve an enzyme of the non-heme Fe²⁺- and αKG-dependent dioxygenase family resembling the formation of (2S,3S)-β-OH-Arg.^19,27,28 Indeed a C_β-hydroxylase SadA of this family has been shown to convert simple N-acylated-l-Leu substrates such as N-succinyl-l-Leu and N-acetyl-l-Leu to the hydroxylated product with (2S,3R)-stereochemical configuration, opposite the C_β stereochemistry found in 3-5.²⁹ Alternatively, hydroxylation could hypothetically be catalyzed by a member of the cytochrome P450 protein family, enzymes that are known to hydroxylate a variety of amino acids found in peptidyl natural products. The formation of (2R,3S)-2 in NRPS-derived skyllamycin has been shown to involve a multi-functional cytochrome P450 (P450_sky) that catalyzes three independent C_β-hydroxylations: hydroxylation of l-Phe, l-O-methyl-l-Tyr, and l-Leu.^23,24 Unlike the other aforementioned hydroxylating catalysts, P450_sky functions on substrates covalently bonded to T domains, which appears to be the emerging paradigm for cytochrome P450-catalyzed hydroxylation of NRPS-derived peptides.³⁰ Consequently, the biosynthesis of the azinothricins, which harbor the identical (2S,3S)-2 stereoisomer that is found in 3-5, has been proposed to occur by the same cytochrome P450- and T domain-dependent C_β-hydroxylation strategy.^31,32 Nonetheless, the mechanism and timing for the biosynthesis of (2S,3S)-2 within natural products, particularly for NRPS-derived peptides, remains entirely speculative.

Here we use feeding experiments and in vitro characterization of recombinant enzymes from the biosynthetic pathways for muraymycin and the chymostatin family to define the mechanism of formation of (2S,3S)-2 in muraymycins along with a new catalytic strategy for generating the capreomycidine ring of (2S,3S)-1. The former involves a non-heme Fe²⁺- and αKG-dependent dioxygenase and, in contrast to other reported C_β-hydroxylations, is demonstrated to occur at a post-NRPS stage of the biosynthesis, i.e., late-stage C_β-H functionalization. The latter involves an FAD-dependent acyl-CoA-like dehydrogenase and occurs during the process of peptide biosynthesis, i.e., a middle-stage C_β-H functionalization, comparable to the timing of the reaction catalyzed by P450_sky. The functional assignment of these two enzyme catalysts from two distinct families of oxidoreductases now affords additional biochemical means for C_β-H functionalization of NRPS-derived natural products.

RESULTS

Isotopic enrichment.

To probe the metabolic origin of the 1 residue of the pseudotripeptide of 3-5, isotopic enrichment studies with l-[D₇,¹⁵N₄]Arg feeding were initially performed. HR-MS of purified 3 revealed a natural abundance ion with an expected mass and isotopic distribution corresponding to 3 (Figure 2a and Figure S1). A second set of peaks was apparent, and examination of the isotopic distribution suggested two species: one peak corresponding to a +9 amu species with 17% enrichment and one corresponding to +10 amu with 7% enrichment. We speculated that l-Arg catabolism, particularly with respect to nitrogen metabolism, might be responsible for observing two separate, isotopically enriched species. Consequently, l-[D₇]Arg was fed to the strain, and HR-MS of 3 revealed the expected (M + H)⁺ ions corresponding to natural abundance 3 and a prominent +5 amu peak corresponding to [D₅]3 (Figure 2a and Figure S1). The isotopic distribution of the +5 amu peak was consistent with a single entity with 37% overall enrichment, a nominal value that is consistent with prior [1-¹³C]-Leu enrichment studies with the 3 producer.³³ The MS/MS fragmentation pattern for muraymycins has been examined in great detail³⁴ and was utilized to identify the site of enrichment. Fragmentation of synthetic (2RS,3RS)-1 and natural abundance 3 yielded (M + H)⁺ ions consistent with the expected mass for the 1 ring fragment ion (Figure 2b and Figure S2). MS/MS analysis of [D₅]3 derived from l-[D₇]Arg feeding (Figure 2b and Figure S2) and [D₅,¹⁵N₄]3 derived from l-[D₇,¹⁵N₄]Arg feeding (Figure S2) gave the expected + 5 and +8 amu peaks, respectively, consistent with all 5 D atoms residing in the 1 ring. Identical results were obtained upon purification and analysis of 4 (Figure S3) and 5 (Figure S4).

Figure 2.

MS analysis of (2S,3S)-1-containing natural products following isotope enrichment with Arg isotopologues. (a) HR-MS of 3 following feeding of the producing strain with l-[D₇,¹⁵N₄]Arg or with l-[D₇]Arg. The expected (M + H)⁺ parent ions are m/z = 930.4533 for 3; m/z = 939.4728 for [D₅,¹⁵N₄]3; m/z = 940.4698 for [D₅,¹⁵N₅]3, and m/z = 935.4846 for [D₅]3. (b) (+)-HR-ESI-MS/MS of natural abundance 3 (top) and [D₅]3 following feeding with l-[D₇]Arg. The expected (M + H)⁺ fragment ions are m/z = 98.0718 for 3 and m/z = 103.1027 for [D₅]3. (c) (+)-HR-ESI-MS of 6 and 7 following feeding with l-[D₇]Arg. The expected (M + H)⁺ parent ions are m/z = 608.3197 for 6; m/z = 613.3510 for [D₅]6; m/z = 594.3040 for 7; and m/z = 599.3354 for [D₅]7. (d) (+)-HR-ESI-MS/MS of natural abundance 7 (top) and [D₅]7 following feeding with l-[D₇]Arg.

The biosynthetic origin of 1 of chymostatins was similarly interrogated with feeding experiments. MS analysis of commercial chymostatin revealed a mixture of chymostatin A (6) and chymostatin B (7), and fragmentation of both yielded the diagnostic ions for the 1 ring (Figure S5). Although the titre was very low, 6 and 7, with the latter produced as the major congener, were detected following fermentation of the producing strain, Streptomyces lavendulae subsp. lavendulae strain NRRL B-2775 (Figure S6). When l-[D₇]Arg was fed, +5 amu peaks corresponding to [D₅]6 and [D₅]7 were detected (Figure 2c and Figure S6). Similar to MS analysis of 3-5, comparison of the fragmentation ions from natural abundance and labeled 7 revealed the 1 ring as the site of enrichment (Figure 2d and Figure S7). Overall, the enrichment data for 3-7 are consistent with the retention of all N atoms and the loss of two H atoms—one from C_α and the other from C_β—during the conversion of l-Arg to (2S,3S)-1. The feeding data are identical with prior reports examining the enrichment of (2R,3S)-1 in faulknamycin,¹⁵ thereby suggesting a hydroxylation-dehydration-conjugate addition mechanism is plausible during the biosynthesis of (2S,3S)-1 in 3-7.

NRPS A domain specificity for muraymycins.

At the onset these studies, only the muraymycin gene cluster was sequenced and available in the public domain (Table S1).³⁵ Consequently, the NRPS system involved in muraymycin biosynthesis was initially targeted for characterization with the primary aim to gain insight into the timing of C_β-H functionalization. NRPSs typically consist of multiple modules, with each module containing minimally 3 domains: adenylation (A), condensation (C), and thiolation (T) that are responsible for the incorporation of one carboxylate-containing unit (typically an amino acid) into a peptide. A domains of NRPSs, which use ATP to activate and load the substrate to the sulfhydryl group of the 4’-phosphopantheine prosthetic group of the adjacent T domain, are generally considered gatekeepers due to their substrate specificity, and structural and biochemical studies have yielded a so-called NRPS code to correlate A domain sequence with specificity.^36,37 For muraymycin, three A domain-containing NRPSs Mur12, Mur21, and Mur27 are apparent from bioinformatic analysis. Mur27 is a single domain protein predicted to activate l-Ile, Mur21 is a didomain protein with a C-terminal T domain and an N-terminal A domain whose substrate specificity could not be predicted, and Mur12 is a C-A-T tridomain protein predicted to activate l-Arg.

Recombinant Mur12, Mur21, and Mur27 were heterologously produced in E. coli (Figure S8). While Mur12 was soluble as a His₆-tagged protein, Mur21 and Mur27 were only obtained in partially pure form as maltose binding protein (MBP) fusions. The A domain specificity was initially examined using the amino acid-dependent ATP-PP_i exchange assay with proteinogenic amino acids. Mur12 preferentially activated l-Arg as predicted from bioinformatics (Figure S9). To further explore the A domain selectivity, hypothetical pathway intermediates (2S,3S)-β-OH-Arg and (2S,3R)-β-OH-Arg were synthesized as reported³⁸ along with the aforementioned (2RS,3RS)-1,³⁹ and Mur12 was tested with these amino acids along with commercial Arg variants including d-Arg and N_α-acetyl-l-Arg. Additionally, commercial (2S,3S)-2, a hypothetical intermediate if hydroxylation of l-Leu occurs before incorporation into the peptide, was examined. The data clearly show that Mur12 has preferential activity with l-Arg when compared with these other amino acids (Figure 3a and Figure S9). Intramolecular thioesterification of the T domain was then analyzed using l-[¹⁴C]Arg, first by incubating Mur12 with CoA and a promiscuous phosphopantetheinyl transferase, Sfp (Figure S8),⁴⁰ and then monitoring the loading by SDS-PAGE with phosphor imaging. A faint band corresponding to the molecular weight of Mur12 was detected, which is consistent with 1 formation occurring after thioesterification (Figure S10). Mur21 showed preferential activity with l-Leu and, importantly, was inactive with commercial (2S,3S)-2 (Figure 3b and Figure S9). Self-loading of l-[¹⁴C]Leu, however, was not observed (Figure S10). Finally, Mur27 had relatively low activity yet with modest specificity toward l-Val (Figure 3c and Figure S9). We hypothesized that Mur14, a T domain protein, was the amino acid acceptor for Mur27 and was therefore produced in E. coli and purified (Figure S8). Similar to the results with Mur21, however, loading of l-[¹⁴C]Val was not observed (Figure S10), both results of which are a potential consequence of interference from the fused MBP or the requirement of an allosteric modulator for A-domain-catalyzed thioesterification.⁴¹ Nonetheless, the data with Mur12 and Mur21 are consistent with the selective activation of the respective proteinogenic amino acids, thus suggesting C_β-H functionalization of l-Arg—unlike during viomycin/capreomycin and faulknamycin biosynthesis—and l-Leu occurs after the A domain-catalyzed reaction.

Figure 3.

Muraymycin NRPS A domain specificity. Relative activity of (a) Mur12, (b) Mur21, and (c) Mur27 with representative amino acids. Counts were processed by subtracting the background prior to calculating the relative activity. The proposed reaction catalyzed by the respective NRPS is shown.

Oxidoreductase I: C_β-hydroxylation of Leu.

Mechanistic precedent for C_β-H functionalization prompted us to search for putative non-heme Fe²⁺- and αKG-dependent dioxygenases encoded in the gene cluster.^11,13,19,27 Two proteins—Mur15 and Mur16—were identified (Table S1). Mur16 was previously functionally assigned as an Fe²⁺- and αKG-dependent UMP dioxygenase, generating uridine-5’-aldehyde that serves as the precursor of both sugars of the nucleoside core (Figure S11).⁴² Past characterization also revealed Mur16 did not hydroxylate l-Arg.⁴² In contrast, VioC, produced here to serve as an experimental control (Figure S8), was shown to hydroxylate free l-Arg (Figure S12). Inclusion of VioD (Figure S8) with VioC in the reaction led to the generation of (2S,3R)-1 as reported (Figure S13),¹¹ and reactions with [D₇]Arg revealed a +5 mass shift for 1 and the respective, diagnostic fragment ions (Figure S14), corresponding to the loss of both H_α and H_β during formation of (2S,3R)-1. Despite no formal loss of hydrogen atoms during the VioD-catalyzed reaction, the data suggest the reaction occurs with exchange of H_α that is likely a consequence of a solvent-exposed active site and/or a mechanism employing a multiprotic Lys as the general acid/base, thus the abstracted D atom would be expected to be “diluted” out by protium (Figure S15). The results of these in vitro labelling results are comparable to the in vivo feeding experiments with the 3-5, 6, and 7 producers (Figure 2), leaving open the possibility for a dehydration-conjugate addition mechanism despite the apparent lack of a VioD homolog encoded within the muraymycin gene cluster.

Mur15 has sequence similarity to proteins annotated as cupin 4 proteins,⁴³ and sequence-based structural predictions using SWISS-MODEL revealed Mur15 has the closest homology to the non-heme Fe²⁺- and αKG-dependent dioxygenase YcfD, a 50S ribosomal protein L15-Arg-β-hydroxylase,⁴⁴ and NO66, a ribosomal protein-His-β-hydroxylase.⁴⁵ Following purification (Figure S8), recombinant Mur15 was tested for potential activity with the free amino acids l-Arg and l-Leu, and the activity was assessed by LC-MS using authentic, hydroxylated l-Arg and l-Leu to facilitate peak identification. However, no peaks corresponding to a hydroxylated product were detected (Figure S12). We also attempted to load l-Arg to the T domain of Mur12 prior to incubating with Mur15, but hydroxylated Arg was similarly not observed (Figure S12). The comparable experiment with Mur21 was not performed since l-[¹⁴C]Leu was not loaded based on SDS-PAGE and phosphor imaging analysis.

Considering the negative results and that Mur15 homologues YcfD and NO66 catalyze post-translational modifications of their respective protein substrates, we next aimed to test activity on a late-stage intermediate, i.e. following amide bond-forming event(s). Unfortunately, the muraymycin production spectrum and relative titres from the wild-type strain were not previously disclosed. Notwithstanding, (2S,3S)-1 is found in all 20+ reported muraymycin congeners produced from the mutant strains. Contrastingly, mutant strain Streptomyces sp. NRRL 30475 appeared to only produce the D series of muraymycins that have l-Leu instead of (2S,3S)-2.^4,5 This finding suggested hydroxylation was impaired in this strain and provided the opportunity to test the fully formed scaffold as a potential substrate for Mur15. Subsequently, 3 was purified from Streptomyces sp. NRRL 30475 (Figure S16) and incubated with Mur15 and the predicted cofactor Fe²⁺ and cosubstrates αKG and O₂ (Figure 4a). LC-MS analysis revealed a new peak with a mass corresponding to mono-hydroxylated 3 (Figure 4b), the formation of which was dependent upon the inclusion of αKG (Figure S17). MS/MS analysis was consistent with Leu as the site of hydroxylation (Figure S18), and the product was ultimately confirmed as muraymycin C1 (8) by co-injections with authentic material isolated from Streptomyces sp. NRRL 30473 (Figure 4b) and spectroscopic analyses (Figure S19 and Figure S20). Therefore, Mur15 is functionally assigned as an αKG-dependent 3-Leu-C_β-hydroxylase (Figure 4a) and represents a rare example of C_β-H functionalization of NRPS-derived natural products that occurs after the peptide backbone is completely assembled.

Figure 4.

Characterization of Mur15. (a) The Mur15-catalyzed reaction. (b) HPLC traces of reactions with 3 and (i) the exclusion of Mur15, (ii) all components, and (iii) all components co-injected with authentic 8. (c) Single-substrate kinetic analysis with variable 3. (d) Structure of the synthetic deaminoribose-3 analogue and HPLC traces of Mur15-catalyzed reactions with (i) the exclusion of Mur15 and (ii) all components.

Biochemical characterization of Mur15 revealed the activity was dependent on O₂, and a single O atom from ¹⁸O₂ was incorporated into the product (Figure S21). The inclusion of Fe²⁺ slightly enhanced the activity with an optimum at 500 μM FeCl₂ (Figure S17). For in vitro activity, many non-heme Fe²⁺- and αKG-dependent dioxygenases depend on ascorbic acid,^19,46 which is used to reversibly reduce an oxidized iron-enzyme inactive state.⁴⁷ However, the addition of ascorbic acid did not significantly alter Mur15 activity (Figure S17). Single-substrate kinetic analysis with saturating αKG revealed Michaelis-Menten kinetics yielding a K_m = 0.6 ± 0.1 mM and k_cat = 10 ± 1 min⁻¹ with respect to 3 (Figure 4c). The total turnover number (TTN) with the native substrate 3 was estimated to be ≥ 800.

To explore the substrate scope and further verify the regioselectivity of hydroxylation, 4 (Figure S22), 8, muraymycin B2 (Figure S23), muraymycin B6 (Figure S24), and muraymycin A1 (Figure S25) were isolated, verified by spectroscopic analysis, and tested with Mur15. Furthermore, a deaminoribose-3 analogue with l-Arg in place of (2S,3S)-1 was synthesized and examined as a substrate (Figure 4d). LC-MS revealed both 4 and the deaminoribose-3 analogue were mono-hydroxylated (Figure 4d and Figure S26), while 8 remained unchanged. MS/MS analysis of the deaminoribose-3 analogue product supported C_β hydroxylation of the Leu residue (Figure S27), although the site was inferred at this time due to a lack of sufficient material for comparative NMR. New peaks were unexpectedly observed with substrates muraymycin B2, B6, and A1 that had a mass consistent with a mono-hydroxylated product (Figure S28). The reaction with muraymycin B2 was used to examine the regiochemistry of hydroxylation. MS/MS analysis of the product was consistent with hydroxylation at the Leu residue (Figure S29), and ¹H-NMR of the product revealed the loss of the Leu-C_γ proton with a downfield shift of the Leu-C_δ protons (Figure S30), indicating that Mur15-catalyzed hydroxylation occurred at the γ position of the Leu moiety to give an unnatural hydroxy-muramycin B2 (Figure S30). Single-substrate kinetic analysis with saturating αKG revealed Michaelis-Menten kinetics yielding a K_m = (9.3 ± 1.3) x 10⁻¹ mM and k_cat = 6.5 ± 0.5 min⁻¹ with respect to muraymycin B2 (Figure S31), corresponding to only a ~2-fold drop in catalytic efficiency when compared with 3. Interestingly, the TNN (≥ 4000) with the unnatural substrate muraymycin B was significantly increased when compared with 3. Finally, inhibition tests demonstrated the important role of hydroxylation on biological activity: the conversion of 3 to 8 improves the IC₅₀ ~30-fold against S. aureus MraY^7,48 but had minimal effect on the anti-M. smegmatis and E. coli activity (Table S3). In contrast, unnatural hydroxylation of muraymycin B2 had no effect on the IC₅₀ or activity against E. coli ΔtolC, but anti-M. smegmatis and E. coli BL21(DE3) activity were abolished.

NRPS A domain specificity for chymostatins and α-MAPI/antipain.

Shortly after the discovery of Mur15 activity, the biosynthetic gene clusters for 6 and 7 from Streptomyces lavendulae subsp. lavendulae strain NRRL B-2775 and the structurally related α-MAPI (9) and antipain (10) from Streptomyces albulus NRRL B-3066 were reported.⁴⁹ In contrast to the chymostatins, both 9 and 10 contain an l-Arg in place of (2S,3S)-1 (Figure 1a). Analysis of the 6/7 gene cluster revealed a unique protein SlAnpI not found in the 9/10 gene cluster (Table S2). Based on sequence SlAnpI is a member of the acyl-CoA dehydrogenase (ACAD) family that typically catalyze FAD-dependent α,β-dehydrogenation of a CoA-thioesterified substrate. As noted in the paper,⁴⁹ the muraymycin gene cluster encodes a protein Mur22 that has 56% sequence identity with SlAnpI. Thus, SlAnpI and Mur22 were proposed to be involved in the formation of (2S,3S)-1.

Toward establishing the role of SlAnpI and Mur22 and timing of (2S,3S)-1 formation, we first partially characterized the NRPSs involved in 6/7 and 9/10 biosynthesis. Despite consisting of four amino acid units, the 6/7 and 9/10 gene clusters encode only three orthologous A domain-containing NRPSs, AnpD, AnpE, and AnpF. Of the three, only the A-domain specificity of AnpF for l-Val could be predicted by sequence analysis. The AnpE and AnpF orthologs were readily obtained upon production in E. coli, while both orthologs of AnpD gave only faint bands corresponding to the predicted molecular weight (Figure S8). Like Mur12, the AnpD orthologs are tridomain C-A-T NRPSs; a truncated version of SlAnpD lacking the C domain was prepared to give a soluble protein that was partially purified. Using the amino acid-dependent ATP-PP_i exchange assay, SlAnpD(AT) had high specificity toward l-Arg (Figure S32). Phosphor imaging of gels following the incubation of holo-SlAnpD(AT) with l-[¹⁴C]Arg clearly revealed the protein was covalently modified (Figure S10). The AnpE orthologs had the highest activity with the aromatic amino acids, l-Leu, and l-Met (Figure S32). The very low activity with l-Val and l-Ile, which substitute for l-Leu in the congeners 7 and chymostatin C (Figure 1b), respectively, suggest that the in vivo substrate for AnpE is likely l-Phe. Consistent with this, covalent modification of the AnpE orthologs was detected following incubation with l-[¹⁴C]Phe and not with l-[¹⁴C]Val (Figure S10). The AnpF orthologs, also predicted A-T didomain proteins, demonstrated low specificity with slight preference for l-Phe (but no activity with l-Tyr and l-Trp), most aliphatic amino acids (l-Leu, l-Val, l-Ile, l-Met), l-His, and l-Cys (Figure S32). Covalent modification of AnpF was observed with l-[¹⁴C]Val but not l-[¹⁴C]Phe (Figure S10), suggesting AnpF is only involved in the incorporation of the central l-Leu, l-Val, or l-Ile residue. The relatively high activity of AnpF with l-Phe using the amino acid-dependent ATP-PP_i exchange assay, however, is nonetheless interesting since the mechanism for adding the second l-Phe found in all chymostatins and 9 prior to reductive off-loading is unknown. Presumably l-Phe is loaded intermolecularly to AnpG, which consists of sequential C, T, and reductase (R) domains. Although likely catalyzed by AnpE, which clearly activates and self-loads l-Phe, data here does not exclude AnpF as this possible catalyst.

Oxidoreductase II: Conversion of l-Arg to (2S,3S)-1.

Recombinant AnpI and Mur22 were produced in E. coli and, despite the relatively high sequence homology (56% identity, 71% similarity) between the two proteins, only AnpI was soluble (Figure S8). The yellow color, UV/Vis spectrum, and HPLC analysis of denatured AnpI revealed that the recombinant protein co-purified with a noncovalently bound FAD (Figure S33). We first examined the potential for AnpI to use l-Arg, but unlike reactions with VioC and VioD (Figure S13), 1 was not detected (Figure S34). This data is not only consistent with the demonstrated l-Arg specificity of the A domain of Mur12/AnpD (Figure S10 and Figure S32) but also with respect to mechanistic considerations (vide infra). To examine the potential for late-stage functionalization, the closest mimics for the completed peptide scaffolds with l-Arg substituting for (2S,3S)-1—the aforementioned deaminoribose-3 analogue for muraymycin and 9 for the chymostatins—were tested with AnpI. Neither, however, were converted to a product (Figure S34), again suggesting that the AnpI-catalyzed reaction occurrs on a carrier protein-bound intermediate during NRPS-mediated assembly. Consequently, the T domain of Mur12 or SlAnpD(AT) was self-loaded with l-Arg and incubated with AnpI, and the hydrolysate analyzed by LCMS (Figure S34). A peak with the same retention time as synthetic (2RS,3RS)-1 was detected in two trials with Mur12, yet the signal was close to the limit of detection. Several attempts to improve the yields and further reproduce the data, including by using l-[D₇]Arg in the reactions as a tracer, were unsuccessful. Based on a standard curve with authentic material, the amount, if indeed (2RS,3RS)-1, was estimated to be 1 nM in the reaction, corresponding to 0.03% yield with respect to the limiting reagent, the T domain of Mur12 (3.4 μM).

Due to the very low yield and irreproducible results, we speculated that l-Phe-(C=O)-l-Arg (11) pseudodipeptide, thioesterified to a T domain, was the true AnpI substrate (Figure 5a). To test this, the production of 11 by the appropriate NRPS components was first confirmed. Thus, Phe-activating SlAnpE was incubated with Mur12, l-Phe, l-Arg, NaHCO₃, and the other necessary components,⁵⁰ and the hydrolysate analyzed by LC-MS (Figure 5a). In contrast to the control omitting Mur12, a new peak appeared with an identical retention time and mass compared to synthetic 11 (Figure 5b and Figure S35). The identity of the new peak was further supported by performing reactions with l-[phenyl-D₅]Phe, NaH¹³CO₃, or l-[D₇]Arg, which gave the product 11 with the respective isotopic shift corresponding to the incorporation of the heavy atoms (Figure 5b). Despite the uncertainty of whether full-length SlAnpD was produced in E. coli (Figure S8), LC-MS revealed the formation of 11 when Mur12 was substituted with SlAnpD (Figure S36). The estimated yield of the reaction was 4.5% with Mur12 as the limiting reagent.

Figure 5.

Characterization of AnpI. (a) Reactions catalyzed by the indicated NRPS and AnpI, and hydrolysis of the product to generate 11 and 12, respectively. (b) HPLC traces detecting 11 from the indicated enzymes using natural abundance l-Phe, l-Arg, and NaHCO₃ (bottom two traces) or substituting the indicated isotopically labelled substrate for the natural abundance counterpart. Data were collected using the indicated mass with m/z expansion set to ± 100 ppm. The limit of detection for 11 was estimated to be 200 fmol based on a signal-to-noise ratio ≥ 3 for the EIC. This value corresponds to a concentration of 20 nM (10 mL injection) from a 50 mL reaction. (c) HPLC traces detecting 12 from the indicated enzymes using natural abundance l-Phe, l-Arg, and NaHCO₃ (bottom two traces) or substituting the indicated isotopically labelled substrate for the natural abundance counterpart. Data were collected using the indicated mass with m/z expansion set to ± 100 ppm. The limit of detection for 12 was estimated to be 200 fmol based on a signal-to-noise ratio ≥ 3 for the EIC. This value corresponds to a concentration of 20 nM (10 mL injection) from a 50 mL reaction.

After confirmation of 11 formation, AnpI was included in the reactions with Mur12 and SlAnpE prior to termination of the reaction by base hydrolysis. A new peak with a mass corresponding to Phe-(C=O)-(2S,3S)-1 (12) was observed at a retention time of ~9.6 min (Figure 5c). To simplify the product identification, 12 was synthesized based on an established protocol for the stereocontrolled preparation of pseudodipeptides containing (2S,3S)-1,⁵¹ and LC-MS analysis revealed the same retention time and MS spectrum (Figure S37). Additionally, when l-[phenyl-D₅]Phe or l-[D₇]Arg was used as a substrate, 12 with the appropriate mass shift was observed (Figure 5c). Importantly, reactions with l-[D₇]Arg gave [D₅]12 (Figure S37), and MS/MS analysis revealed a +5 amu in the 1 ring fragment ion (Figure S38), both of which are consistent with the feeding results. Although complete conversion was not achieved, [11]:[12] ratios typically ranged between 2.2:1-1:1, corresponding to yields of 1.4-2.3% with respect to Mur12 (Figure S39). No product was observed when AnpI was incubated with synthetic 11 (Figure S34), suggesting that thioesterification is necessary for AnpI catalysis. Finally, both termini of 11 are carboxylic acids and, in theory, either could be thioesterified to the respective T domain following the formation of the ureido bond and prior to the AnpI-catalyzed reaction. To ascertain which amino acid residue is thioesterified following condensation, hydrolysis of the AnpD/Mur12 and AnpE reaction product was performed with K¹⁸OH. MS analysis revealed the expected [¹⁸O₁]11 product (Figure S40), and MS/MS revealed a fragmentation pattern consistent with the incorporation of ¹⁸O into the carboxylate of the l-Arg terminus as shown in Figure 5a. Thus, the data are consistent with the functional assignment of AnpI as an FAD-dependent l-Phe-(C=O)-l-Arg-S-AnpD (or Mur12) dehydrogenase:cyclase that generates the (2S,3S)-1 ring.

DISCUSSION

Several natural product biosynthetic proteins highlighted by two new oxidoreductases involved in C_β-H functionalization of amino acid residues of NRPS-derived muraymycins and chymostatins have been functionally characterized. The results are consistent with biosynthetic pathways wherein C_β-H functionalization of both l-Arg and l-Leu residues occurs after N-acylation to form ureido and amide bonds, respectively (Scheme 1). The modification of the former amino acid, however, occurs earlier in the biosynthesis. In our proposed pathway, the A domain of an NRPS, either Mur12 for muraymycin or AnpD for the chymostatin family, activates and loads l-Arg via thioesterification to the T domain to generate l-Arg-S-Mur12/AnpD. In parallel, Mur27 or AnpE activate and load the “upstream” amino acid, l-Val and l-Phe, respectively, to distinct T domains. While AnpE was directly shown to catalyze intramolecular loading, Mur27 consists solely of an A domain, and thus we surmise that Mur14, a free-standing T domain that is the only remaining unassigned T domain encoded in the 3-5 biosynthetic gene cluster, is loaded by Mur27. However, intermolecular loading to generate l-Val-S-Mur14 could not be demonstrated for as-of-yet unknown reasons. Following uriedo bond formation between the two amino acid units, MS/MS data from ¹⁸O-hydrolysis (Figure S39) clearly show that the remaining thioester bond is between the arginyl residue and Mur12/AnpD. Thus, it can be inferred that the A domains of Mur27 and AnpE also adenylate and condense the bicarbonate unit to the amine of l-Val-S-Mur14 and l-Phe-S-AnpE, respectively, prior to ureido bond formation. Past efforts with SylC, a C-A-T tridomain NRPS that forms a ureido bond between two l-Val residues during syringolin A biosynthesis, provided evidence that bicarbonate is likely incorporated by one of two proposed cyclization mechanisms.⁵⁰ Due to our inability to radiolabel any NRPS with [¹⁴C]NaHCO₃ (Figure S10), a diffusible anhydride is depicted as the hypothetical cyclized intermediate in Scheme 1. Regardless of the nature of intermediate, the C domain of AnpD is proposed to catalyze ureido bond formation to generate 11-S-AnpD or, in the case of muraymycin biosynthesis, Mur12-catalyzed formation of l-Val-(C=O)-l-Arg-S-Mur12. Interestingly, the Mur12-catalyzed condensation event appears nonspecific with respect to the acyl donor since l-Phe incorporation was readily accomplished in place of the natural l-Val residue. Further studies will be needed to fully define the biochemical role of these NRPSs and their catalytic domains in ureido bond formation.

Scheme 1.

Biosynthesis of the peptide component of muraymycins and chymostatins and the two oxidative transformations characterized in this study.

Once 11-S-Mur12/AnpD is formed, the ACAD homolog AnpI—and by extension Mur22 for muraymycin biosynthesis—catalyzes FAD-dependent dehydrogenation to form an α,β-unsaturated thioester that serves as the immediate precursor for N_η attack by nucleophilic-conjugate addition. We propose that AnpI/Mur22 also catalyzes this conjugate addition, although the 6-exo-trig cyclization could potentially be mediated by the Mur12/AnpD C domain analogous to the proposal put forth for faulknamycin.¹⁵ Regardless, AnpI/Mur22 now joins KtzA, which uses a thioesterified γ-chlorinated l-Ile to catalyze cyclopropyl formation without concomitant FAD reduction (Figure S41),⁵² and PltE/RedW, which uses thioesterified l-Pro to oxidize the pyrrolidine ring to ultimately form a pyrrole,⁵³ as in vitro characterized ACAD homologs that work in concert with an NRPS to incorporate a nonproteinogenic amino acid into the peptide scaffold. Several more biosynthetic gene clusters for NRPS-derived natural products encode for ACADs that putatively catalyze distinct reactions on amino acid residues, yet remain to be characterized.^54–56 Moreover, it is highly likely that other heterocyclic amino acids, not just 1, are generated in a similar manner. For example, the gene cluster for tambromycin, which contains the unusual 5-membered cyclic amino acid tambroline that is derived from l-Lys, encodes for minimally two ACAD homologs (TbrP and TbrQ) that could potentially catalyze dehydrogenation-conjugate addition .^57,58 Similarly, the gene cluster for azinomycin B encodes multiple ACAD homologs (AziC8, AziD2, and AziD3),⁵⁹ one or more of which are likely involved in the biosynthesis of the unusual bicyclic amino acid, an aziridno[1,2a]pyrrolidinyl residue, that originates from l-Glu.⁶⁰

The general accepted mechanism for ACADs involves a concerted C_α–H deprotonation with β-hydride elimination.⁶¹ Several studies with model systems in aqueous solutions have shown that thioesterification of (amino) acids dramatically lowers the C_α-H pK_as in aqueous solution when compared to the free carboxylate counterpart.⁶² From a mechanistic point of view, it is thus not surprising that AnpI/Mur22, like all other known ACADs,⁶¹ utilizes a thioesterified substrate. The same trend, however, does not hold when comparing the free, cationic amine (NH₃⁺) versus the N-acylated amino acid counterpart, wherein there are minimal differences in the estimated C_α-H pK_as.⁶³ Therefore, the timing of the AnpI/Mur22-catalyzed reaction with respect to the N-acylation could not be readily predicted a priori. Our data are consistent with the AnpI/Mur22-catalyzed reaction preferentially occurring with the N-acylated substrate, which is unqiue when compared to other members of the ACAD superfamily. Further biochemical and structural studies with AnpI/Mur22 are now needed to ascertain the role of the N-acylation, if any, in substrate recognition and catalysis for this newest family member. Nonetheless, once 12-S-Mur12/AnpD (or the l-Val-containing pseudodisaccharide variant) is formed, the NRPS-guided assembly is proposed to continue with the condensation of the downstream amino acid (Scheme 1). For chymostatins, an aliphatic amino acid is activated by AnpF, and we propose AnpC catalyzes the condensation between the two thioesterified substrates. Further downstream processing includes adding a second l-Phe, proposed to be catalyzed in part by AnpE, and reductive off-loading catalyzed by AnpG. For muraymycins, this amino acid is l-Leu, despite most congeners having (2S,3S)-2 at this position. Mur21 activates and loads l-Leu, and we propose that the C domain protein Mur13 catalyzes the subsequent condensation. Off-loading of the pseudotripeptide likely occurs with the mature nucleoside core as an acyl acceptor, potentially in a reaction catalyzed by a C domain-like protein Mur25.

While we initially hypothesized that Mur15 hydroxylated l-Arg due to the biosynthetic precedent for 1 formation and the sequence similarity of Mur15 to l-Arg-β-hydroxylases, this changed following our on-going investigation marred with negative results. Although Mur15 was ultimately shown to catalyze C_β hydroxylation, this modification occurs on the Leu residue of muraymycins. Of the Fe²⁺- and αKG-dependent dioxygenases that hydroxylate l-Leu, l-Leu-5-hydroxylase, which is involved in the biosynthesis of 4-methyl-Pro⁶⁴, and LdoA, a l-Leu-5-hydroxylase that has an unknown biological function,⁶⁵ both utilize the free amino acid l-Leu. Likewise, SadA, which catalyzes the comparable C_β-hydroxylation of l-Leu that Mur15 performs although with opposite stereochemistry, uses an N-acylated substrate.²⁹ SadA, like LdoA, has an unknown biological function. On the other hand, BcmC, one of five Fe²⁺- and αKG-dependent dioxygenases involved in bicyclomycin biosynthesis, uses an Ile-Leu diketopiperazine substrate to catalyze hydroxylation of Leu at yet another position, Cγ.⁶⁶ Interestingly, a different Fe²⁺- and αKG-dependent enzyme BcmB was shown to modify C_β of a Leu residue that is part of an Ile-Leu diketopiperazine substrate, but in this case does not function as a hydroxylase but instead generates a C_α,C_β-epoxide. However, none of these examples are linked to the biosynthesis of an NRPS-derived natural product. Contrastingly, P450_sky, which is involved in the biosynthesis of an NRPS-derived natural product skyllamycin, utilizes a thioesterified l-Leu (l-Leu-S-PCP₁₁) to afford a C_β-hydroxylated product with ~ 2% conversion to product, which is incidentally similar in yield to the thioester-dependent AnpI-catalyzed reaction. By functioning on the completed peptidyl nucleoside scaffold, Mur15 invokes a distinct substrate selection modality than P450_sky, SadA, and the aforementioned other Leu hydroxylases to achieve β-hydroxylation and is henceforth defined as a post-NRPS tailoring enzyme. It remains to be seen if the biosynthesis of other, (2S,3S)-2-containing NRPS-derived peptides follow such a course.^{20,21,25,31,67–70}

The discovery of the function of Mur15 now establishes a new biocatalytic tool for generating novel 2-containing muraymycin analogues, proof of principle of which was demonstrated with a deaminoribose-3 analogue. The biocatalytic potential was further expanded by the unexpected discovery of Mur15-catalyzed Cγ hydroxylation of C_β-O-acylated muraymycin congeners. This moonlighting activity is notable since the reaction occurs at a sterically dense, tertiary carbon. Hydroxylation at such tertiary centers of amino acids is known, including C_β-hydroxylation of Val and C_γ-hydroxylation of Leu. Notably, both modifications can be achieved by Fe²⁺- and αKG-dependent enzymes, the former represented by PoyI involved in the biosynthesis of the ribosomally encoded polytheonamides⁷¹ and the latter represented by the previously mentioned BcmC.⁶⁶ The rather serendipitous discovery of the two discrete hydroxylations catalyzed by Mur15, creating either a secondary or tertiary alcohol, continues to add to the ever-expanding inventory of this enzyme superfamily that is reputed for their catalytic power and chemical diversity, highlighted in part by these unique-yet-related group of Leu-modifying enzymes.

In summary, the NRPS system involved in the biosynthesis of muraymycins and the chymostatin family have been interrogated to reveal the proteinogenic amino acids are the likely immediate substrates. Two distinct mechanisms for C_β-H functionalization have been described, one (Mur15) employing a traditional hydroxylation that results in C_β-O bond formation, and the other (AnpI/Mur22) an unusual dehydrogenation that results in C_β-N bond formation. Both of these modifications occur after minimally one NRPS-mediated condensation event, with the former occurring after the scaffold has completely assembled and hence can appropriately be considered a late-stage functionalization. Given the general importance of C_β-H functionalization with respect to biological activity, the uncovering of these enzymes now adds to the biocatalytic armament that could potentially be used for the synthesis and discovery of new entities of biologic utility or for other applications.

ABBREVIATIONS

αKG

α-ketoglutarate

MS

mass spectrometry

NMR

nuclear magnetic resonance

NRPS

nonribosomal peptide synthetase

A

adenylation

C

condensation

T

thiolation