Bacterial strains and plasmids used in the current study are described in Table 1. All media/reagents were purchased from Fisher Scientific or VWR International (Canada). For routine analysis, Streptomyces cultures were maintained on International Streptomyces Project (ISP) medium 4 agar or were grown in Trypticase Soy Broth supplemented with 1% (w/v) soluble starch (TSB-S). Cultures for metabolite analysis were grown using glycerol, sucrose, proline, and glutamic acid (GSPG); starch asparagine (SA); soy; or TSB-S media (Romero et al., 1997; Tahlan et al., 2004b). All Streptomyces cultures were incubated at 28°C and liquid cultures were agitated at 250 rpm. Plasmid-bearing Streptomyces cultures were supplemented with appropriate antibiotics when required (Tahlan et al., 2004b), whereas Escherichia coli was grown and maintained as described previously (Sambrook, 2001). Standard methods were used for isolating and manipulating DNA from E. coli (Sambrook, 2001) and Streptomyces (Kieser et al., 2000; Tahlan et al., 2004b). Total RNA was isolated from S. clavuligerus grown on SA medium as described previously (Srivastava et al., 2019), and RT was performed using the Maxima H Minus First Strand cDNA Synthesis Kit (Thermo Scientific, United States). PCRs were carried out using the Phusion or Taq DNA polymerase kits (ThermoFisher, United States). When required, PCR products were cloned into the pGEM-T Easy vector (Promega, United States) according to the manufacturer’s instruction and were sequenced at the Centre for Applied Genomics, University of Toronto (Canada). All DNA oligonucleotide primers used in the study (Supplementary Table S2) were obtained from Integrated DNA Technologies (United States).

The S. jumonjinensis and S. katsurahamanus genomes were sequenced using Illumina MiSEQ in paired-end format with read lengths of 300 bp. A chromosomal DNA library was prepared for each organism using the PCR-based method adjusted for high GC DNA according to the manufacturer’s instructions (Illumina, United States). Raw reads were filtered with trimmomatic (Bolger et al., 2014) with a cutoff of 26 bp and a minimum length of 150 bp. The remaining reads were assembled using Velvet (Zerbino and Birney, 2008). k-mers from 30 to 170 were tested for selecting optimal contig length and the assembled genomes (31–46 × coverage, Supplementary Table S3) were submitted to NCBI (accession numbers: S. jumonjinensis NRRL 5741, VCLA00000000 and S. katsurahamanus T-272, VDEQ00000000). Genome completeness was calculated (Supplementary Table S3) using BUSCO (Simao et al., 2015) and QUAST (Gurevich et al., 2013). Annotations were carried out using RAST (Overbeek et al., 2014) and also manually in Artemis (Rutherford et al., 2000). Specialized metabolite (SM) biosynthetic gene clusters (BGCs) were identified using antiSMASH 4.0 (Blin et al., 2017) and polyketide synthases/nonribosomal peptide synthetase genes were predicted using PRISM 3 (Skinnider et al., 2017). The DNA sequences of S. jumonjinensis and S. katsurahamanus BGCs were manually examined for possible frame shifts and other ambiguities. In some cases, PCR amplification was performed using custom primers (Supplementary Table S2) followed by Sanger sequencing of products to verify results. The genome sequences of S. clavuligerus ATCC 27064 (NZ_CM000913.1, NZ_CM000914.1), S. pratensis ATCC 33331 (NC_016114), S. viridis DSM 43017 (CP001683.1), Streptomyces sp. M41(2017) (NZ_MWFK00000000), Streptomyces sp. PAMC26508 (NC_021055), Streptomyces sp. NRRL S-325 (NZ_JOIW00000000), Streptomyces sp. NRRL B-24051 (NZ_JOAE00000000), Streptomyces flavovirens NRRL B-2182 (NZ_JOAB00000000), Streptomyces fulvoviridis NRRL ISP-5210 (NZ_JNXH00000000), and Streptomyces olivaceus NRRL B-3009 (NZ_JOFH00000000) were included for comparison as the latter harbor CA-like BGCs containing homologs of all genes currently known to be involved in CA production in S. clavuligerus (Jensen, 2012). In addition, the sequences of the Ceph-C BGCs from Streptomyces cattleya 8057 (NC_017586.1) and Nocardia lactamdurans (also known as Amycolatopsis lactamdurans) (Z13971.1-Z13974.1, Z21681.1-Z21686.1 and X57310.1) were also included in the analysis. Geneious 8.1.9 (Biomatters Ltd., New Zealand) was used for sequence comparisons and constructing phylogenetic trees. Protein homologs were identified using NCBI BLAST and secretory signals were predicted using the SignalP-5.0 Server (Almagro Armenteros et al., 2019).

The S. clavuligerus nocE gene mutant was prepared using the meganuclease I-SceI marker-less gene deletion system (Fernández-Martínez and Bibb, 2014). DNA fragments (1–1.2 kb each) containing regions immediately upstream and downstream of nocE from the S. clavuligerus chromosome were amplified using PCR along with engineered primers (Supplementary Table S2) and were separately cloned into the pGEM-T Easy vector (Table 1). The upstream fragment was released from pGEM-T Easy by digestion with HindIII and EcoRI and was introduced into the same sites of pIJ12738 to give pIJ12738/nocE-UP. The downstream fragment was then introduced into the EcoRI and XbaI sites of pIJ12738-nocE-UP to give pIJ12738/nocE-UP-DN, which functioned as the nocE disruption construct (Table 1). pIJ12738-nocE-UP-DN was conjugated into S. clavuligerus to obtain the apramycin-resistant single crossover strain, which was confirmed using genomic DNA PCR (Supplementary Table S2). The plasmid pIJ12742 expressing the I-SceI meganuclease (Table 1) was then conjugated into S. clavuligerus pIJ12738-nocE-UP-DN to obtain apramycin and thiostrepton resistant exconjugants, which were made to undergo sporulation at 28°C without any selection to facilitate double homologous recombination and loss of pIJ12738 from the chromosome. Spore stocks were prepared and re-streaked onto ISP-4 plates without selection and incubated for 5 days at 37°C to promote the loss of temperature-sensitive pIJ12742. This led to the isolation of the apramycin and thiostrepton-sensitive S. clavuligerus ΔnocE mutant, which was verified using genomic DNA PCR (Supplementary Table S2).

To prepare an S. clavuligerus strain constitutively expressing nocE (Table 1), the ermEp^∗ promoter (Bibb et al., 1985) was inserted upstream of the gene in the S. clavuligerus chromosome. A 1.1-kb DNA fragment from the 5′ end of the gene was amplified by PCR (Supplementary Table S2) and was cloned into pGEM-T Easy. The insert was re-isolated as an NdeI and EcoRI fragment and was ligated with similarly digested pIJ8668-ermEp^∗ to give pIJ8668-ermEp^∗-nocE (Table 1), which was introduced into wt S. clavuligerus by conjugation. This resulted in the S. clavuligerus ermEp^∗-nocE strain, which was confirmed using genomic DNA PCR (Supplementary Table S2) and was used to examine the effect of constitutively expressing nocE in S. clavuligerus.

The production of CA and Ceph-C in culture supernatants was detected (and quantified in the case of Ceph-C) using bioassays employing indicator organisms (Table 1), as described previously (Paradkar and Jensen, 1995; Wang et al., 2004). Growth in liquid cultures was determined using a modified diphenylamine colorimetric method to measure DNA content (Zhao et al., 2013) and statistical analysis (ANOVA repeated measure) was performed using R 3.4.3. To assess for growth characteristics on solid media, 10-fold dilutions of a spore stock (4 × 10⁴ spores/μl) were prepared, and 5 μl of which were spotted onto two different agar media (SA and TSB-S with 1.5% agar). The plates were then incubated at 28°C and visually scored for growth over a 7-day period.

The production of clavam metabolites in 96-h broth cultures was analyzed by targeted LC-MS after imidazole derivatization using an XTerra column (2.1 × 150 mm, 3.5 μm, 125 Å; Waters Scientific, United States) as described previously (Srivastava et al., 2019). Untargeted metabolomics was conducted using bacteria grown on solid media. One hundred microliters of a standardized spore stock (4 × 10⁴ spores/μl) of each species was used to inoculate agar plates in duplicate, and each plate was extracted using 15 ml of methanol or ethyl acetate. Two milliliters of each extract was dried, resuspended in 130 μl of 70% methanol containing 0.2 μM of amitriptyline (internal standard), and transferred to a 96-well plate, which was centrifuged at 2000 rpm for 15 min at 4°C. One hundred microliters of each sample was then transferred to a new 96-well plate for LC-MS/MS analysis. Samples were analyzed using a Vanquish UHPLC System coupled Q Exactive Hybrid Quadrupole-Orbitrap Mass Spectrometer (Thermo Scientific, United States). Chromatographic separation was performed in mixed mode (allowing weak anion/cation exchange) on a Scherzo SM-C18 column (2 × 250 mm, 3 μm, 130 Å; Imtakt, United States) maintained at 40°C. Ten microliters of each sample was injected for analysis and the mobile phase consisted of (A) 0.1% formic acid in water and (B) 0.1% formic acid in acetonitrile. Chromatography was performed at a flow rate of 0.5 ml/min using the following program: 0–5 min, 98% A; 5–8 min, gradient of 98–50% A (or 50% B); 8–13 min, gradient 50–100% B; 13–14.00 min, 100% B; 14–14.10 min, 100–2% B; 14.10–18 min, 2% B.

Mass spectrometry was performed using a heated electrospray ionization source (heater temperature, 370°C and capillary temperature, 350°C) in either positive or negative ionization mode (± 3000.0 V; S-lens RF, 55; sheath gas flow rate, 55; and auxiliary gas flow rate, 20). MS¹ and MS² scans (at 200 m/z) were acquired from 0.48 to 16.0 min at a resolution of 35,000 and 17,500, respectively, for the 100–1500 m/z range. The automatic gain control (AGC) target value and maximum injection time were set at 5 × 10⁵ and 150 ms. Up to four MS² scans in data-dependent mode were acquired for most abundant ions per duty cycle, with a starting value of 70 m/z, and exclusion parameter of 10 s. Higher-energy collision-induced dissociation was performed with a normalized collision energy of 20, 35, and 50 eV. The apex trigger mode was used (2–7 s) and the isotopes were excluded. Inclusion lists of ions for molecules observed in Streptomyces extracts were generated from the Dictionary of Natural Products¹ and the StreptomeDB (Lucas et al., 2013), and were used for prioritizing the acquisition of their MS² when observed. The raw LC-MS/MS data files were converted to .mzXML format using ProteoWizard (Adusumilli and Mallick, 2017). All metabolomics MS data have been deposited on the MassIVE public repository² under the accession number MSV000083835.

Molecular networks were generated using positive and negative ionization mode data in GNPS (Wang et al., 2016). The resulting networks were visualized in Cytoscape (Shannon et al., 2003), allowing nodes associated with uninoculated media controls to be removed. Annotations were first obtained by matching spectra in public libraries (Wang et al., 2016), including NIST17³. Library annotations were manually validated using mirror plots (maximum ion mass accuracy = 5 ppm) corresponding to level 2 annotation based on the Minimum Standard Initiative (Spicer et al., 2017). The data were deposited to the GNPS library (CCMSLIB00005435954-CCMSLIB00000531493), which enabled the annotation of putative tunicamycin derivatives (CCMSLIB00005435941-42) and lyngbyatoxin (CCMSLIB00005435954-55) using molecular networks. In some cases, Sirius 4.0.1 was used to confirm the molecular formulas of certain predicted metabolites (Böcker et al., 2009).

To generate a heat map using the S. clavuligerus wt, ΔnocE, and ermEp^∗-nocE strains, feature-based detection and alignment of positive mode ionization data were performed (parameters: MS¹ noise level of 25000, MS² noise level of 1000) using the MZmine 2 toolbox (v2.39) (Pluskal et al., 2010). Chromatograms were built using the ADAP module (parameters: min group size in # of scans = 4, group intensity threshold = 700,000, min highest intensity = 100,000, max m/z tolerance = 10 ppm), which were then deconvoluted (parameters: S/N threshold = 10.0, min feature height = 7000000, coefficient/area threshold = 60.0, peak duration range = 0.01–0.5 min, RT wavelet range = 0.01–0.1 s). Fragmentation spectra were paired with deconvoluted peaks using 0.02 Da and 0.2 min windows, and LC-MS features were annotated using the Peak-Grouping module (parameters: deisotope = true, remove features without isotope pattern = false, minimal intensity for interval selection = 0.1, minimal intensity overlap = 0.7, minimal correlation = 0.7). Features were aligned in the JoinAligner module (parameters: ppm tolerance = 7, weight for m/z = 75.0, retention time tolerance = 0.5 min, weight for RT = 25.0; require same charge state = false, require same ID = false, compare isotope pattern = false). The aligned peaklist was filtered with the row filter module to keep only features with at least two isotopic ions, two occurrences, and at least one MS² spectrum before gap filling (parameters: intensity = 5%, ppm window = 5, retention time tolerance = 0.15). The aligned peaklist containing 3149 features was exported as a .CSV file, and the spectral data as .MGF files using the GNPSExport module for further processing. The signal intensities of the features (.CSV) were normalized to that of an internal standard (m/z 278.189; retention time, 9.2 min) and only 1684 features with an intensity 3-fold higher than in experimental controls (uncultivated media) were retained. MetaboAnalyst4.0 (Chong et al., 2018) was used to perform the hierarchal clustering, which was visualized as a heat map.

Detailed analysis of the S. jumonjinensis and S. katsurahamanus genome sequences using antiSMASH 4.0 (Blin et al., 2017) and manual curation showed that both organisms contain numerous BGCs for diverse SMs (Figure 2A and Supplementary Table S4). Therefore, S. clavuligerus, S. jumonjinensis, and S. katsurahamanus were grown on SA, GSPG, and TSB-S media for assessing CA/Ceph-C production (Figure 2B and Supplementary Figure S1) and for preparing methanol/ethyl acetate extracts for liquid chromatography–tandem mass spectrometry (LC-MS/MS)-based metabolomics. The MS/MS data obtained from both positive and negative ionization mode were used to build a molecular network (Figure 2C), and metabolites were annotated by matching spectra against public libraries corresponding to level 2 annotation based on the Metabolomics Standard Initiative (Wang et al., 2016). During the analysis, ions corresponding to CA ([M-H]^–, m/z 198.039), Ceph-C ([M-H]^–, m/z 445.104) and numerous other SMs were also detected in extracts from one or more Streptomyces species (Figure 2C, Supplementary Tables S5, S6), some of which are discussed below.

The desferrioxamines (Figure 2C) comprise a group of nonpeptide hydroxamate siderophores produced by many bacteria (Barona-Gómez et al., 2004), including S. clavuligerus (Álvarez-Álvarez et al., 2017). In the current study, ions corresponding to desferrioxamine E (Nocardamine, [M + H]⁺, m/z 601.356) and desferrioxamine B (Desferal, [M + H]⁺, m/z 561.361) were detected in extracts from S. clavuligerus/S. jumonjinensis and S. clavuligerus/S. katsurahamanus, respectively (Supplementary Table S5). Desferrioxamine E exhibits antitumor activity (Kalinovskaya et al., 2011), while desferrioxamine B is used in therapy for secondary iron overload disease (Olivieri and Brittenham, 1997). We also identified BGCs in S. clavuligerus, S. jumonjinensis, and S. katsurahamanus (Supplementary Table S4) that have high degrees of similarity (80–100%) with BGCs from known desferrioxamine producers such as Streptomyces griseus (Yamanaka et al., 2005; Ohnishi et al., 2008) and Streptomyces coelicolor A3(2) (Bentley et al., 2002; Barona-Gómez et al., 2004). The siderophore arthrobactin (Figure 2C) was also detected in S. katsurahamanus extracts ([M + H]⁺, m/z 477.256) (Supplementary Table S5), but since the genes responsible for arthrobactin production are not known (Burrell et al., 2012), we were unable to identify an associated BGC in this organism. However, our analysis showed that S. clavuligerus, S. jumonjinensis, and S. katsurahamanus each contain additional siderophore-like BGCs of unknown function (Supplementary Table S4), which could potentially be involved in the production of such metabolites. Ectoine is another commonly produced metabolite that helps bacteria survive extreme osmotic stress (Sadeghi et al., 2014), and it was detected ([M + H]⁺, m/z 143.082) in extracts from all three CA-producing species (Figure 2C and Supplementary Table S5). In addition, S. clavuligerus, S. jumonjinensis, and S. katsurahamanus contain BGCs that are similar to the known ectoine BGC from Streptomyces anulatus (previously called Streptomyces chrysomallus) (Prabhu et al., 2004). Since the desferrioxamines and ectoine are produced by many Actinomycetes and are involved in general cellular growth/survival processes (Challis, 2005; Czech et al., 2018), finding them in culture extracts from the three CA producers in the current study was not surprising.

Streptomyces clavuligerus is a known producer of the dithiolopyrrolone antibiotic holomycin (Kenig and Reading, 1979) and the associated BGC has been identified in this organism (Li and Walsh, 2010). In the current study, holomycin ([M + H]⁺, m/z 214.994) and thiolutin (another dithiolopyrrolone, [M + H]⁺, m/z 229.010) were detected in extracts from S. clavuligerus, but not in those from S. jumonjinensis or S. katsurahamanus (Figure 2C and Supplementary Table S5). Recently, a dithiolopyrrolone with the same molecular weight as thiolutin (predicted to be N-propionylholothin) was also detected in extracts from S. clavuligerus strains lacking the giant linear plasmid pSCL4 (Álvarez-Álvarez et al., 2017). Since holomycin and thiolutin (Figure 2C), and the respective BGCs involved in their biosynthesis (from S. clavuligerus and Saccharothrix algeriensis NRRL B-24137, respectively), are very similar (Supplementary Table S4), it is possible that a single pathway in S. clavuligerus produces both metabolites. It has also been reported that there is some sort of cross regulation between CA and holomycin production in S. clavuligerus (de la Fuente et al., 2002; Álvarez-Álvarez et al., 2017). Our results showed that S. jumonjinensis and S. katsurahamanus lack dithiolopyrrolone BGCs (Supplementary Table S4) and therefore do not have a similar link between holomycin and CA production as observed in S. clavuligerus.

We also detected certain nucleoside SMs during the current analysis (Figure 2C). For example, the purine nucleoside pentostatin, which is also used as an anticancer agent (Dillman, 2004), was identified ([M + 2H]²⁺, m/z 135.066) in S. clavuligerus extracts (Figure 2C, Supplementary Table S5). A putative pentostatin-like BGC was recently shown to be present in S. clavuligerus (Wu et al., 2017), but production of the metabolite has not been reported in this organism previously. Therefore, our results suggest that the S. clavuligerus pentostatin BGC can be activated under laboratory conditions. The tunicamycins also comprise a mixture of related nucleoside antibiotics, some of which (A, B, C, and I) were detected in extracts from S. clavuligerus (Figure 2C and Supplementary Table S5), but not in those from S. jumonjinensis or S. katsurahamanus. S. clavuligerus is a known producer of tunicamycin and the BGC involved in its production has been identified (Kenig and Reading, 1979; Chen et al., 2010). In addition, certain derivatives of tunicamycin I with different acyl chains were detected in S. clavuligerus extracts recently (Martínez-Burgo et al., 2019), which were also present in our samples (Supplementary Table S5). Our results demonstrated that S. jumonjinensis and S. katsurahamanus do not possess tunicamycin BGCs (Supplementary Table S4), further distinguishing S. clavuligerus from the other CA producers due to its ability to produce such nucleoside SMs.

Metabolomics analysis also revealed the presence of certain plant-associated SMs in the Streptomyces extracts. It was recently shown that S. clavuligerus produces the citrus flavonoid naringenin and the genes involved in the production of this metabolite were also identified (Álvarez-Álvarez et al., 2015). Naringenin exhibits antibacterial, antifungal, and anticancer activities (Rauha et al., 2000; Kanno et al., 2005), and its production by a bacterium was unexpected since it was previously isolated from plants only (Álvarez-Álvarez et al., 2015). We detected naringenin (Figure 2C, [M-H]^–, m/z 271.062) in extracts from S. clavuligerus and S. jumonjinensis, but not from S. katsurahamanus (Supplementary Table S5). In addition, the genes involved in naringenin production were also found in both S. jumonjinensis and S. katsurahamanus (Supplementary Table S4), suggesting that the metabolite might be produced at undetectable levels in S. katsurahamanus or that the genes are not expressed in this species under the conditions tested. Also detected in all three Streptomyces extracts were the plant-associated monoterpenes, carveol ([M-H₂O + H]⁺, m/z 135.117), and cuminyl alcohol ([M-H₂O + H]⁺, m/z 133.101), whereas hydroxyvalerenic acid (another plant terpene, [2M-H]^–, m/z 499.307) was found in S. clavuligerus extracts only (Figure 2C and Supplementary Table S5). The pathways involved in the production of the latter three metabolites are not fully known (Wong et al., 2018), however, S. clavuligerus, S. jumonjinensis, and S. katsurahamanus possess many terpene-like BGCs of unknown function, which could potentially be involved in their biosynthesis (Supplementary Table S4). Therefore, our results suggest that certain Streptomyces also harbor the capacity to produce carveol, cuminyl alcohol, and hydroxyvalerenic acid along with naringenin, a finding that can be potentially exploited for further development.

The indole alkaloid, (−)-indolactam V is a protein kinase C activator (Heikkila and Akerman, 1989) and functions as an intermediate during the biosynthesis of other SMs in certain Actinomycetes (Abe, 2018). We detected (−)-indolactam V (Figure 2C, [M-CO + H]⁺, m/z 274.191) and some of its alkylated derivatives in extracts from S. clavuligerus, but not in those from S. jumonjinensis or S. katsurahamanus (Supplementary Table S5). The genes normally associated with (−)-indolactam V biosynthesis could not be identified in the current study, warranting further investigation into its production in S. clavuligerus. Other metabolites were also detected during the analysis (Supplementary Table S6), but we were unable to find details about their biosynthesis in bacteria or predict associated BGCs, and therefore we did not include them in the discussion. In addition, S. jumonjinensis and S. katsurahamanus contain several BGCs related to known pathways for which products could not be detected (Supplementary Table S4). For example, there is an NRPS-containing BGC in S. jumonjinensis that is 100% similar to the BGC in Streptomyces sp. DSM 11171, which produces the antiviral metabolite feglymycin (Supplementary Table S4; Gonsior et al., 2015). We also identified indole-associated BGCs in S. clavuligerus and S. jumonjinensis (Supplementary Table S4), which are similar to the one from Streptomyces sp. TP-A0274 responsible for producing the anticancer agent staurosporine (Onaka et al., 2002). Similarly, BGCs for polycyclic tetramate macrolactams (PTMs, NRP/PKs) are present in both S. jumonjinensis and S. katsurahamanus, which are 100% similar to a SGR-PTM BGC from the known producer S. griseus (Supplementary Table S4; Luo et al., 2013). PTMs possess antifungal and antioxidant properties, and cryptic PTM-like BGCs are commonly found in Streptomyces genomes (Zhang et al., 2016). Moreover, BGCs for many other classes of SMs including enediynes (Rudolf et al., 2016) and the ribosomally synthesized and post-translationally modified peptides (RiPPs) (Hetrick and van der Donk, 2017) were also identified in S. jumonjinensis and S. katsurahamanus (Supplementary Table S4), but further work is required to detect their production in these organisms. In the current study, >14,000 molecular nodes were obtained using MS-based metabolomics and GNPS analysis (Figure 2C), but only 10% could be annotated by matching spectra with available libraries. Therefore, many of the unannotated nodes could represent products of so-called “cryptic” BGCs, a situation that should change over time as databases are populated with more spectra from authentic samples.

In addition to analyzing the overall SM production capabilities of CA producers, we were also interested in specifically examining the BGCs involved in β-lactam biosynthesis from S. jumonjinensis and S. katsurahamanus for comparison with S. clavuligerus (Figure 3). The genome sequences of S. jumonjinensis and S. katsurahamanus revealed that they both contain identical CA and Ceph-C BGCs (Figure 3), but lack the clavam and paralog gene clusters (Supplementary Table S4). This would explain why they do not produce the 5S clavams as compared to S. clavuligerus (Jensen, 2012). The results further confirm that intact 5S clavam and paralog BGCs are not essential for CA production (Figure 1), since both S. jumonjinensis and S. katsurahamanus can produce the metabolite (Figure 2B and Supplementary Figure S1). The paralog gene cluster from S. clavuligerus contains second copies of certain genes (ceaS1, bls1, and pah1) from the CA BGC (Jensen et al., 2004b; Tahlan et al., 2004b), which encode enzymes involved in the early shared stages of CA and 5S clavam biosynthesis (Figure 1). It has also been shown that the remaining un-duplicated genes from the paralog gene cluster and almost all genes from the clavam gene cluster (except one; cas1) are exclusively involved in 5S clavam production (Mosher et al., 1999; Tahlan et al., 2007; Zelyas et al., 2008). Therefore, our results provide additional support for the hypothesis that the clavam and paralog gene clusters are associated with 5S clavam biosynthesis, and that some gene products from the two clusters augment CA production in S. clavuligerus by contributing to a common pool of precursors (Figure 1; Jensen, 2012; Hamed et al., 2013). Although, it should be noted that in S. clavuligerus, there is some cross regulation between the chromosomal CA and plasmid-borne paralog gene clusters (Kwong et al., 2013; Álvarez-Álvarez et al., 2017), which is again not expected to occur in the other two CA producers since they only contain the CA BGC. This also highlights the complexity of the regulatory pathways controlling CA and 5S clavam production in S. clavuligerus (Liras et al., 2008). For this reason, we focused our analysis and discussion on the comparison of biosynthetic genes (and BGCs), instead of regulation. In the current study, CA production levels in S. jumonjinensis and S. katsurahamanus could never match those observed in wt S. clavuligerus, whereas all three species produced Ceph-C at comparable levels (Figure 2B). It has been previously suggested that higher CA yields in S. clavuligerus might be explained in part by increased precursor supply for biosynthesis due to the presence of the paralog and clavam gene clusters in this species (Figure 1). In addition, enhanced levels of biosynthetic gene expression could be another reason why S. clavuligerus is currently the preferred industrial producer and was first identified in screens for β-lactamase inhibitors, as higher CA yields would make it easier to detect during assays (Jensen, 2012).

Closer examination of the CA BGCs from S. jumonjinensis and S. katsurahamanus showed that they each contain most of the genes from the corresponding S. clavuligerus BGC in the same order, except that orf18 (pbpA), orf20, orf21, orf22, and orf23 are absent (Figure 3A). pbpA is predicted to encode a high-molecular-weight penicillin-binding protein (PBP), but its role in CA production remains unknown (Jensen et al., 2004a). Previous studies have also shown that disruption of orf19 (pbp2) (Jensen et al., 2004a), orf20 (cytochrome P-450) (Song et al., 2009), orf21 (putative sigma factor), orf22 (sensor kinase), or orf23 (response regulator) (Fu et al., 2019a) in S. clavuligerus does not abolish CA or Ceph-C production (Song et al., 2009; Supplementary Table S1). Since the respective genes are not present in S. jumonjinensis and S. katsurahamanus (Figure 3A), it is apparent that they are not part of the core BGC required for biosynthesis, but instead have accessory roles in S. clavuligerus. In a previous study, it was also shown that the expression of orf18–21 was not significantly affected in a S. clavuligerus mutant defective in ClaR, the cluster-situated regulator responsible for controlling CA biosynthesis (Martínez-Burgo et al., 2015). Therefore, we propose that the core CA BGC comprises ceaS2 (encoding carboxyethylarginine synthase), gcas (encoding N-glycyl-clavaminic acid synthetase), and the intervening genes (Figure 3A, and Supplementary Table S1).

The CA and Ceph-C BGCs in S. jumonjinensis and S. katsurahamanus also form “β-lactam superclusters” as observed in S. clavuligerus, which agrees with previous restriction mapping studies (Ward and Hodgson, 1993). The linkage of the Ceph-C and CA BGCs in S. clavuligerus, S. jumonjinensis, and S. katsurahamanus, and the coordinated production of the two metabolites in S. clavuligerus (Pérez-Llarena et al., 1997), provides further evidence for the simultaneous acquisition of the two BGCs by producing species. It has been proposed that the CA BGC might have evolved by the duplication of an ancestral 5S clavam BGC and the acquisition of the ability to produce Ceph-C in the same organism (Challis and Hopwood, 2003). Such a situation led to the selection for the ability to produce a β-lactamase inhibitor, resulting in the assembly of the currently known CA BGC, and the formation of the β-lactam supercluster (Challis and Hopwood, 2003). Our results showed that the Ceph-C BGCs from S. jumonjinensis and S. katsurahamanus are identical to each other, but differ slightly from those present in S. clavuligerus and other Ceph-C-producing Actinobacteria (Figures 3B,D). The positions of genes forming individual operons (or transcriptional units) in all three CA producers is very similar (except for the location of cefD), but the relative arrangement of operons is different in S. jumonjinensis and S. katsurahamanus as compared to S. clavuligerus (Figure 3B). In addition, the Ceph-C BGCs of S. jumonjinensis, S. katsurahamanus and other previously reported Ceph-C producers (other than S. clavuligerus) (Liras et al., 1998) do not contain blp (Figure 3B), which encodes a product resembling β-lactamase inhibitory proteins (Blip), but has been shown to lack any such activity (Gretes et al., 2009). Previous studies have shown that disruption of blp does not affect Ceph-C or CA production in S. clavuligerus (Alexander and Jensen, 1998; Thai et al., 2001). Therefore, blp does not seem to a part of the core Ceph-C BGC since S. jumonjinensis, S. katsurahamanus, and other species shown in Figure 3B can still produce the metabolite in its absence. Another noticeable feature of Ceph-C BGCs from the three CA producers is the presence of pcbR, which is missing from the homologous BGCs of species that only produce Ceph-C, but not CA (Figure 3B). PcbR resembles PBPs (Paradkar et al., 1996), but it is not essential for Ceph-C biosynthesis since it is not present in the BGCs of all organisms capable of producing the metabolite (Figure 3B, more details below).

Overall, the “β-lactam superclusters” from S. clavuligerus, S. jumonjinensis, and S. katsurahamanus are very similar to each other (Figures 3C,D). In comparison, CA-like BGCs from non-producers are markedly different, and do not form “β-lactam superclusters” as they lack Ceph-C BGCs (Jensen, 2012). The non-producers (including some Streptomyces) are also phylogenetically distinct from CA-producing species (Supplementary Figure S2), and their CA-like BGCs show three distinct patterns in terms of gene content and arrangement (Figure 3A). Many organisms in the database contain CA-like BGCs identical to the one found in S. pratensis, whereas we could only find one example each of the types present in Streptomyces sp. M41 and S. viridis, respectively (Figure 3A). In addition, CA-like BGCs from S. pratensis and S. viridis contain the pcbR, orf11, and nocE genes (Álvarez-Álvarez et al., 2013), which are not present in the CA BGCs of S. clavuligerus, S. jumonjinensis, or S. katsurahamanus (Figure 3A). Interestingly, pcbR and orf11 are included in the Ceph-C BGCs of CA producers, whereas nocE is located elsewhere on the chromosome in the three Streptomyces species (Figure 3B). As mentioned earlier, pcbR encodes a PBP involved in β-lactam resistance (Paradkar et al., 1996), whereas orf11 encodes a predicted protein of unknown function. Previous reports have shown that disruption of neither pcbR nor orf11 in S. clavuligerus affected Ceph-C or CA production (Paradkar et al., 1996; Alexander and Jensen, 1998), suggesting that they are not required for the biosynthesis of the respective metabolites.

The presence of nocE homologs in CA producers and in the CA-like BGCs of all non-producers is intriguing (Figure 3A), as they are similar to a gene from the nocardicin A monobactam BGC of Nocardia uniformis (Gunsior et al., 2004). The nocE genes are predicted to encode proteins containing C-terminal SGNH/GDSL hydrolase family domains, which are normally associated with esterases or lipases (Upton and Buckley, 1995), but their function during β-lactam metabolite biosynthesis is not obvious. The disruption of nocE in N. uniformis does not affect nocardicin A production (Davidsen and Townsend, 2009), but the role of the gene in β-lactam-producing Streptomyces has not been examined to date.

In previous studies, every gene from the proposed CA BGC of S. clavuligerus (Figure 3A) was systematically disrupted (Supplementary Table S1), to determine if it had any effect on CA or Ceph-C production. It has been suggested that nocE might have some role during CA biosynthesis in S. clavuligerus, but since the gene is not part of the CA BGC, a mutant has not been prepared and analyzed to date (Jensen, 2012). Therefore, the function of nocE was examined in the model CA producer, S. clavuligerus. RT-PCR analysis of RNA isolated from wt S. clavuligerus grown in SA medium demonstrated that nocE is temporally expressed along with ceaS2 and cas2 (Figure 4A), genes that are essential for CA biosynthesis (Figure 1). However, when S. clavuligerus strains were prepared in which nocE was either deleted (ΔnocE) or constitutively expressed (ermEp^∗-nocE) (Table 1), the production of CA, 5S clavams, or Ceph-C was found to be unaffected (Figure 4B and Supplementary Figure S3), demonstrating that the gene is not required for β-lactam metabolite production in S. clavuligerus. The predicted lipase/esterase-like domain present in NocE is also found in hydrolytic enzymes from other Streptomyces species, some of which are known to be secreted (Wei et al., 1995; Vujaklija et al., 2002). Closer examination of the predicted NocE amino acid sequence from S. clavuligerus suggested that it is also a secreted protein, as it contains a highly conserved N-terminal Sec-signal sequence (p > 0.9) (Almagro Armenteros et al., 2019). These findings further ruled out the direct involvement of NocE in CA production, which occurs in the cytoplasm, and suggested that NocE might have some other exocellular hydrolytic function instead. Therefore, the S. clavuligerus wt, ΔnocE, and ermEp^∗-nocE strains were assessed for growth under different nutritional condition using TSB-S (rich), soy (complex fermentation), or SA (defined fermentation) media (Figure 4C). It was observed that the growth of the S. clavuligerusΔnocE mutant was significantly reduced in each medium tested, whereas that of the ermEp^∗-nocE strain was enhanced in SA medium only, when compared to the wt strain (Figure 4C). The growth of the three strains was also assessed on TSB-S and SA agar, which again showed that the S. clavuligerus ΔnocE mutant did not grow as well as the other strains in the latter medium (Figure 4D). To examine the influence of nocE on primary metabolism in S. clavuligerus, the wt, ΔnocE, and ermEp^∗-nocE strains were grown on TSB-S and SA agar for metabolomics analysis, which showed marked differences in overall metabolite levels between the respective strains (Figure 4E). Furthermore, metabolomics analysis showed that SM production in S. clavuligerus was unaffected in the ΔnocE mutant as compared to the wt strain. Therefore, based on all evidence collected so far, it seems plausible that NocE could have some extracellular role in nutrient acquisition in S. clavuligerus, but like pcbR and orf11, it is not required for CA or Ceph-C production under the tested conditions.