Discovery of actinomycin L, a new member of the actinomycin family of antibiotics

Streptomycetes are major producers of bioactive natural products, including the majority of the naturally produced antibiotics. While much of the low-hanging fruit has been discovered, it is predicted that less than 5% of the chemical space of natural products has been mined. Here, we describe the discovery of the novel actinomycins L1 and L2 produced by Streptomyces sp. MBT27, via application of metabolic analysis and molecular networking. Actinomycins L1 and L2 are diastereomers, and the structure of actinomycin L2 was resolved using NMR and single crystal X-ray crystallography. Actinomycin L is formed via spirolinkage of anthranilamide to the 4-oxoproline moiety of actinomycin X2, prior to the condensation of the actinomycin halves. Such a structural feature has not previously been identified in naturally occurring actinomycins. Adding anthranilamide to cultures of the actinomycin X2 producer Streptomyces antibioticus, which has the same biosynthetic gene cluster as Streptomyces sp. MBT27, resulted in the production of actinomycin L. This supports a biosynthetic pathway whereby actinomycin L is produced from two distinct metabolic routes, namely those for actinomycin X2 and for anthranilamide. Actinomycins L1 and L2 showed significant antimicrobial activity against Gram-positive bacteria. Our work shows how new molecules can still be identified even in the oldest of natural product families.


Results
The influence of carbon sources on bioactivity and actinomycin production. Streptomyces sp.
MBT27 is a gifted natural product producer that was isolated from Qinling mountains in China, with potent antibacterial activity against various MDR (multi-drug resistant) bacteria 28 . We previously showed that the strain among others produces the novel quinazolinones A and B 29 . To investigate the antibiotic activity of Streptomyces sp. MBT27 the strain was fermented in minimal medium (MM) with either of the following carbon sources (percentages in w/v): 1% of both mannitol and glycerol, 1% mannitol, 2% mannitol, 1% glycerol, 2% glycerol, 1% glucose, 2% glucose, 1% fructose, 1% arabinose, or 1% N-acetylglucosamine (GlcNAc). Supernatants of Streptomyces sp. MBT27 cultures were extracted with ethyl acetate and bioactivity assays were performed against Bacillus subtilis 168. Interestingly, the carbon sources had a huge effect on the antimicrobial activity (Fig. S2). Particularly strong antimicrobial activity was observed when the culture medium was supplemented with glycerol + mannitol, glucose 1%, glycerol, fructose or GlcNAc; as compared to when mannitol or arabinose were used as the carbon sources.
In order to investigate the metabolic differences due to nutritional supplementation and correlate that to the antimicrobial activity, LC-MS-based metabolomics was performed. Initially, the LC-MS data were explored by unsupervised Principal Component Analysis (PCA). The first two PCs accounted for 37% and 16%, respectively, of the total data variation. PCA analysis failed to show significant metabolic separation in relation to the observed bioactivity (Fig. 1a). The supervised Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) was then applied to discriminate the samples based on their ability to inhibit B. subtilis (Fig. 1b). The cross-validation metrics of the model (R 2 Y = 0.748 and Q 2 Y = 0.676) indicated that the model has a good reliability and ability of prediction. A permutation test was performed (n = 100) and the resulting R 2 Y and Q 2 Y values were significantly lower (p values < 0.01 for both), which indicated that there was no overfitting in the model 30 (Fig. S3). The OPLS-DA loadings S-plot revealed the most discriminative features between active and inactive groups (Fig. 1c).

Figure 1.
Differential production of metabolites depending on the carbon source. (a) PCA score plot of Streptomyces sp. MBT27 metabolites produced in cultures with different carbon sources, namely, 1% arabinose, 1% fructose, 1% GlcNAc, 1% glucose, 2% glucose, 1% glycerol, 2% glycerol, 1% mannitol, 1% mannitol + 1% glycerol and 2% mannitol (%ages in w/v). (b) OPLS-DA score plot. Triangles and crosses represent samples of active and inactive groups respectively, circular areas represent the 95% confidence region of each group. (c) OPLS-DA loadings S-plot. Arrows indicate the most discriminative features that positively correlate with the active groups. Global Natural Product Social (GNPS) molecular networking 23 was subsequently employed to detect MS/ MS-based structural relatedness among features in an automated manner. The web-based platform generates a molecular network wherein features with related scaffolds cluster together. Cytoscape 3.7.2 was used for visualization of the generated molecular networks 32 . A network representing the ions detected in the crude extract of Streptomyces sp. MBT27 grown with 1% glycerol was constructed, revealing 172 nodes clustered in 10 spectral families (Fig. 2). The molecular network revealed an actinomycin spectral family containing actinomycin D, X 2 and X 0β . Moreover, the same spectral family included a yet unidentified compound with m/z 1387.67. It was closely connected (cosine score > 0.7) to the known actinomycins, suggesting that the molecule was a novel actinomycin. Statistical analysis showed that the extracts with stronger antimicrobial activity contained higher concentrations of actinomycins X 2, X 0β , D and the new compound, in comparison with the less active ones (ANOVA, p < 0.05; Fig. 2). It is important to note that actinomycins were only detected in the bioactive extracts. Cultures were grown for seven days in MM with 1% glycerol. Orange nodes represent ions of the metabolites produced by Streptomyces sp. MBT27, while blue nodes represent those of the media components. The actinomycin spectral family is enlarged. Results of ANOVA statistical analysis were mapped onto the molecular network to illustrate the differential production of actinomycin cluster members under various growth conditions. Box plots represent relative intensities of actinomycins X 2 , X 0β , and D after log transformation and pareto scaling; together with a compound with an m/z value of 1387.67, in cultures grown in MM with the following carbon sources: 1. 1% arabinose; 2. 1% fructose; 3. 1% GlcNAc; 4. 1% glucose; 5. 2% glucose; 6. 1% glycerol; 7. 2% glycerol; 8. 1% mannitol; 9. 1% mannitol + 1% glycerol; 10. 2% mannitol (%ages in w/v). www.nature.com/scientificreports/ Large scale fermentation and NMR. To allow identification and structural analysis of the likely novel actinomycin analogue, we performed large-scale fermentation of Streptomyces sp. MBT27 followed by bioactivity guided fractionation. The purification process resulted in the isolation of two compounds (1) and (2), with the same mass (Figs. S4, S5). The NMR spectra of the two compounds were very similar, suggesting that they were diastereomers (Figs. S6-S12). Based on 1D and 2D NMR analysis of 1, together with the molecular formula and degrees of unsaturation dictated by the accurate mass, the structure of the isolated diastereomers was determined as a variant of actinomycin D, whereby an aminal was formed between the amino group of anthranilamide moiety and keto group at the γ position of the proline residues (Fig. 3). The prolyl substitution position is the same as that of the hydroxyl and keto groups in actinomycins X 0β and X 2 , respectively. The new actinomycin analogue was designated actinomycin L (with L standing for Leiden, the city of its discovery). The second stereoisomer (2) was crystallized successfully. Single-crystal X-ray diffraction confirmed the structure obtained for (1) based on NMR, and established the absolute configuration to be 2′S, 2″S, 4′R, 4″R, 10′R, 12′S, 12″S, 18′S, 18″S, 23′R, 23″R by anomalous-dispersion effects in diffraction measurements on the crystal (Fig. 4). As the absolute configuration of the amino acid residues in 2 was consistent with that of previously reported actinomycins 33 , and considering that the two isomers stemmed from the aminal formation at C-10′, compound 1 is inevitably the 10′S isomer of actinomycin L.
Biosynthesis of actinomycin L. Actinomycins D (or X 1 ), X 2 , and X 0β detected in the extracts of Streptomyces sp. MBT27 are members of the actinomycin X complex. Recently it was shown that actinomycins X 0β and X 2 are formed through the sequential oxidation of the γ-prolyl carbon by the cytochrome P450 enzyme saAcmM 34,35 . Based on its structure, actinomycin L is most likely formed through an aminalization reaction between the two amino groups of anthranilamide and the γ-keto group on the proline residue of actinomycin X 2 . Accordingly, its production should be arrested when one of the precursors is not available. Interestingly, Streptomyces sp. MBT27 produced actinomycin L in very low amounts when grown with fructose (1% w/v) as the sole carbon source. Moreover, ANOVA statistical analysis showed that anthranilamide was produced in equally low amounts under the same growth conditions (ANOVA, p < 0.05; Fig. 5). Under conditions where Streptomyces sp. MBT27 produced actinomycin L, namely when grown in MM with 1% GlcNAc, 1% glucose, 1% glycerol, 2% glycerol or in 1% mannitol + 1% glycerol, the strain invariably produced both actinomycin X 2 and anthranilamide. However, under conditions where actinomycin X 2 was produced but not anthranilamide, the strain failed to produce actinomycin L (Fig. 5).
We therefore wondered if anthranilamide may be a precursor for the biosynthesis of actinomycin L. To test this hypothesis, we performed a feeding experiment, whereby anthranilamide was added to cultures of Streptomyces sp. MBT27 grown in MM with 1% fructose, where virtually no actinomycin L was produced. Analysis of the supernatant of the cultures via LC-MS revealed that actinomycin L was readily produced when anthranilamide was added, but not without it (Fig. 6a). This strongly suggested that anthranilamide is required for the production of actinomycin L. However, extracts of Streptomyces sp. MBT27 fermented with 1% fructose and additional anthranilic acid contained both anthranilamide and actinomycin L (Fig. 6a). This suggests that indeed anthranilic acid is converted into anthranilamide, which in turn is incorporated into actinomycin L.
In order to unambiguously verify that actinomycin L was the product of anthranilamide and actinomycin X 2 , we conducted another biotransformation experiment, now feeding anthranilamide to S. antibioticus IMRU 3720, which is a known producer of actinomycins X 2 and X 0β , but fails to produce actinomycin L under any condition tested. In line with our hypothesis, S. antibioticus IMRU 3720 also failed to produce anthranilamide under any of the growth conditions (Fig. S17). Excitingly, LC-MS analysis revealed the production of actinomycin L by S. antibioticus IMRU 3720 when anthranilamide was fed to the cultures, but never without anthranilamide (Fig. 6b). This validates the concept that anthranilamide is a key precursor of actinomycin L. Conversely, when anthranilic acid instead of anthranilamide was added to cultures of S. antibioticus IMRU 3720, we failed to detect actinomycin L and anthranilamide (Fig. 6b).
The oxidation of the proline residue in actinomycins X 0β and X 2 occurs following the formation of the two halves of actinomycin, known as 4-MHA PPLs, and prior to the condensation of these halves to form  Table S1). The exact mass and fragmentation pattern of PPL 3 was consistent with a 4-MHB containing PPL wherein an anthranilamide moiety had been attached to the proline residue (Fig. S21). Taken together, the feeding experiments convincingly show that actinomycin L is formed through reaction of anthranilamide with the 4-keto group on the proline residue in the pentapeptide lactone. Moreover, results of the feeding experiments with 3-hydroxy-4-methylbenzoic acid show that this reaction occurs prior to the condensation of the pentapeptide lactones into actinomycin L (Fig. 7).
Identification of the actinomycin BGC in Streptomyces sp. MBT27. To characterize the BGC responsible for actinomycin biosynthesis and compare the genes with those found in known actinomycin BGCs, Streptomyces sp. MBT27 was sequenced using the PacBio platform. Assembly of the PacBio reads resulted in two contigs of 8.4 Mb and 0.13 Mb in length. Analysis using AntiSMASH 6 37 readily identified the actinomycin BGC in the 8.4 Mb contig. Comparison to the actinomycin X 2 BGC from S. antibioticus showed that all genes were highly conserved between the two clusters (Table S1 and Fig. S22). This strongly suggests that the actinomycin BGC does not specify the observed modifications in the actinomycin structure, and is not responsible for the

Bioactivity of isolated compounds (MIC).
Bioactivity assays were carried out for the actinomycins, to test their ability to act as antibiotics. As expected, the compounds showed selective antibacterial activity against Gram-positive pathogens, and none of the actinomycins presented any activity against E. coli ATCC25922 or K. pneumoniae ATCC700603 (Table 1). All compounds except actinomycin X 0ß showed antibacterial activity against Gram-positive bacteria with MIC values ranging from 1 to 16 µg/mL. Actinomycin L 1 showed somewhat higher bioactivity than actinomycin L 2 , while both compounds showed slightly higher MICs than actinomycin D and Actinomycin X 2 .

Discussion
Actinomycin was the first antibiotic identified in Actinobacteria 24 . The well-established actinomycin structure is composed of a heterocyclic chromophore and two cyclic PPLs. Biosynthetically, PPL is biosynthesized by an NRPS assembly line with the 4-MHA as the initiating unit 38,39 . 4-MHA is derived from 3-hydroxy-4-methylkynurenine (4-MHK), which is formed by methylation of 3-hydroxykynurenine (3-HK) 26 . Our work surprisingly revealed a novel structure within the extensively studied actinomycin family, namely that of actinomycin L, which arises via attachment of an anthranilamide moiety to the γ-carbon of one of the proline residues through aminal formation. ANOVA statistical analysis proved that production of anthranilamide is the limiting factor in the biosynthesis of actinomycin L. Feeding experiments with anthranilamide suggested that actinomycin L is formed through the spontaneous reaction of anthranilamide with the 4-oxoproline site of actinomycin X 2 prior to the condensation of the two 4-MHA PPLs into actinomycin L. To the best of our knowledge, the attachment of anthranilamide to a 4-oxoproline moiety is a novel observation. www.nature.com/scientificreports/ The actinomycin BGC of Streptomyces sp. MBT27 harbors the same genes as that of S. antibioticus, with high homology between the genes, which strongly suggests that the modification of actinomycin X 2 to actinomycin L is not encoded by the BGC itself. Indeed, we anticipate that anthranilamide is derived from anthranilic acid in Streptomyces sp. MBT27, whereby anthranilic acid in turn is biosynthesized through the shikimate pathway 40 . Anthranilic acid is a commonly produced primary metabolite in Streptomyces, while anthranilamide is less common [41][42][43] . The actinomycin X 2 producer S. antibioticus IMRU 3720 fails to convert anthranilic acid into anthranilamide, which explains why actinomycin L was also not detected in the extracts. However, actinomycin L was produced when we fed cultures of S. antibioticus IMRU 3720 with additional anthranilamide, which is fully in line with our proposed biosynthetic pathway. Thus, actinomycin L is an example of a natural product that requires the joining of two separate metabolic pathways, and this is a concept that needs more attention. After all, scientists rely increasingly on heterologous expression and synthetic biology approaches 44 , and these will likely fail if genes are required that do not fall within the main BGC.
The production of actinomycins by Streptomyces spp. is strongly influenced by the carbon source, whereby the preferred carbon source varies from strain to strain [45][46][47][48][49] . D-galactose favors actinomycin production in Streptomyces antibioticus over arabinose, xylose, glucose, fructose and rhamnose 45 , while glycerol was the optimal carbon source for actinomycin production by S. antibioticus Tü 6040 and S. antibioticus SR15.4 46,47 . In the case of Streptomyces sp. MBT27, growth on MM with glycerol, GlcNAc, fructose and glucose (all 1% w/v) as sole carbon sources were the best carbon sources to promote the production of actinomycins. However, increasing the glucose concentration to 2% blocked the production of actinomycins. Glucose was previously reported to repress the transcription of the gene for hydroxykynureninase, which is involved in the formation of the main actinomycin precursor 4-MHA 45 . Importantly, in our experiments the carbon source not only promoted the overall production levels, but also contributed to the chemical diversity of the actinomycins, including the production of actinomycin L. This coincided with the production of anthranilamide, an essential substrate to form this novel actinomycin variant.
In the twenty-first century, genome mining and renewed drug discovery efforts have revealed that Actinobacteria may produce many more molecules than was expected 50 . What is important to note is that this also applies to well-known families of molecules and in extensively studied model organisms. Examples are the highly rearranged cryptic polyketide lugdunomycin that belongs to the family of angucyclines 51 , the new glycopeptide corbomycin with a novel mode of action 52 , as well as the discovery of coelimycin 53 and a novel branch of the actinorhodin biosynthetic pathway 54 in the model organism Streptomyces coelicolor. The discovery of actinomycin  www.nature.com/scientificreports/ L provides another interesting example that we have not yet exhausted the known part of the chemical space. Indeed, the isolation of these novel actinomycins underlines that the biosynthetic potential of Actinobacteria still has major surprises in store, and that we can expect that new molecules can be discovered even within extensively studied microbes and compound classes.

Methods
General experimental procedures. Optical rotation, FT-IR and UV were measured as previously described 55 . NMR spectra were recorded on a Bruker Ascend 850 MHz NMR spectrometer (Bruker BioSpin GmbH). Data was analyzed using MestReNova 14 software (Mestrelab Research, Santiago de Compostela, Spain). The structures of molecules were drawn using ChemDraw Professional version 16.0 (Perkin-Elmer Informatics). HPLC purification was performed on Waters preparative HPLC system as described 29 . All solvents and chemicals were of HPLC or LC-MS grade, depending on the experiment.
Bacterial strains, growth conditions and metabolite extraction. Streptomyces sp. MBT27 was obtained from the Leiden University strain collection and had previously been isolated from the Qingling Mountains, Shanxi province, China 28 . Cultures were grown in triplicate in 100 mL Erlenmeyer flasks containing 30 mL of liquid minimal medium (MM) 56 , supplemented with various carbon sources, and inoculated with 10 µL of 10 9 /mL spore suspension. The carbon sources (percentages in w/v) were: 1% mannitol + 1% glycerol, 1% mannitol, 2% mannitol, 1% glycerol, 2% glycerol, 1% glucose, 2% glucose, 1% fructose, 1% arabinose or 1% N-acetylglucosamine (GlcNAc Up-scale fermentation, extraction and fractionation. Large-scale fermentation, extraction end fractionation were performed as previously described 29 . The fractions eluted with n-hexane-acetone (1:1) was subjected to a SunFire C 18 column (10 μm, 100 Å, 19 × 150 mm) eluted with a H 2 O-MeOH gradient of 50-100% in 20 min, at a flow rate of 15 mL/min. The fraction containing the actinomycins was collected and further purified on semi-preparative SunFire C 18 column (5 μm, 100 Å, 10 × 250 mm), run at 3 mL/min, and eluted using a H 2 O-MeOH gradient of 70-100% in 20 min, to yield actinomycins L 1 (1, 2.9 mg), L 2 (2, 1.3 mg), X 2 (3, 1 mg), X 0β (4, 2.9 mg), and D (5, 3.2 mg). Actinomycin L 1 (  60 . Each compound was serially diluted in DMSO with a dilution factor of 2 to test 10 concentrations starting at 128 μg/mL in all the antimicrobial assays. The MIC was defined as the lowest concentration of compound that inhibited ≥ 95% of the growth of a microorganism after overnight incubation. The Genedata Screener software (Genedata, Inc., Basel, Switzerland) was used to process and analyze the data and to calculate the RZ' factor in the assay that was between 0.90 and 0.98 supporting its robustness.  29 . LC-MS/MS acquisition of the pure compounds was performed using Shimadzu Nexera X2 UHPLC system coupled to Shimadzu 9030 QTOF mass spectrometer as previously described 61 . LC-MS/MS acquisition for molecular networking was performed using Thermo Instruments MS system (LTQ Orbitrap XL, Bremen, Germany) equipped with an electrospray ionization source (ESI) as described 29 .
Computation of mass spectral networks. MS/MS raw data were converted to a 32-bit mzXML file using MSConvert (ProteoWizard) 62 and spectral networks were assembled using Global Natural Product Social molecular networking (GNPS) (https:// gnps. ucsd. edu) as described 23 . Briefly, the precursor ion mass tolerance was set to 2.0 Da and a MS/MS fragment ion tolerance of 0.5 Da, while the minimum cosine score was set to 0.7. The data were clustered using MSCluster with a minimum cluster size of three spectra. The spectra in the