The BRA006, from MicroMarin collection (https://www.labbmar.ufc.br/micromarinbr) was cultured in A1 medium [Starch (10 g/L); Yeast extract (4 g/L); Peptone (2 g/L); Sea Water 75%] in a volume of 100 mL in 250 mL Erlenmeyer flasks. The cultures were centrifuged at 12.000 × g for 10 min and the cell pellets were resuspended in 10 uL of lysozyme 0.05 g/mL for 500 ul of SET buffer. The mixture was incubated at 37°C for 30 min. Then 14 ul Proteinase K (20 mg/mL) and 60 uL SDS 10% were added to the cell lysate and incubated for a further 1 h at 55°C. Then 200 uL NaCl 5 M was added and the temperature was raised to 37°C. 500uL of chloroform was added and the system was centrifuged at 4,500 × g for 10min at 20°C. 500uL was collected in a new tube and 300uL of isopropanol was added and incubated overnight (16 h). The system was centrifuged at 14,000 × g for 10 min at 4°C. The supernatant was discarded and the DNA pellet was resuspended in TE buffer. DNA libraries were prepared using the Oxford Nanopore Ligation Sequencing Kit (SQK-LSK110) and library loading and sequencing were performed according to the manufacturer’s instructions and protocol for Flongle.

For MinION data in use an in-house pipeline with tools suggested by Oxford Nanopore. The acquisition of the raw data, a series of processing steps were followed until these genomes were assembled. The first stage of data processing consisted of converting the raw signals into DNA base sequences (base calling) using the Guppy tool (Wick et al., 2019) and Dorado with super high accuracy settings. With this data in hand, the quality of the readings obtained in the previous step was assessed using the NanoStat software (De Coster et al., 2018). With these results, the processing continued with the removal of adapters from the base called reads to prepare them for subsequent steps; this was done using the Porechop tool (https://github.com/rrwick/Porechop), which also compressed the resulting reads. Once the adapters had been removed, a new quality analysis was carried out using the NanoStat tool. Once this step was completed, the low-quality bases and short reads were trimmed using the Chopper tool (De Coster and Rademakers, 2023), which not only trimmed but also compressed the resulting reads. The next step was to analyze the quality of the resulting sequence, again using the NanoStat tool. With the resulting data, we then used the Flye (Kolmogorov et al., 2019) and Unycicler (Wick et al., 2017a) tools to assemble the genomes using filtered, high-quality reads. The genome assembly was then refined using information from reads mapped using the Racon tool (Vaser et al., 2017). Finally, with the final result in hand, the quality of the final genome assembly was assessed using the Quast tool (Gurevich et al., 2013) and BUSCO (Manni et al., 2021). For Illumina procedures, BRA006 genome samples were sequenced using MiSeq technology at Macrogen facility (Seul, South Korea), all the processing steps such as read mapping, trimming low quality reads and de novo genome assembly were performed using the proprietary software Geneious (Version 11) (Kearse et al., 2012).

For phylogenetic inference, we choose a comprehensive method called digital DNA:DNA hybridization (dDDH) available in the Type Genome Server (TYGS) web tool (Meier-Kolthoff and Göker, 2019). We used both MinION and Illumina data as query sequences with a standard parameter set.

To reveal biosynthetic gene clusters (BGC) from the BRA006 genome, we used the antiSMASH tool (version 7.1.0) (Blin et al., 2023) with relaxed detection strictness and all extra features selection. Coding sequences (CDS) prediction of BRA006 assemblies were made using Prokka (1.14.6) (Seemann, 2014), which is based on prodigal (Hyatt et al., 2010) HMM models to identify proteins by their family motifs. Finally, we combine Prokka and antiSMASH results to obtain a better resolution of protein functions within BGCs. We manually filtered antiSMASH “Most Similar Known Cluster” feature and retrieved BGCs that matched MiBiG clusters related to functions of interest (version 3.0) (Terlouw et al., 2023). Thus, python scripts were used to convert CDS of GenBank files from MiBiG and antiSMASH into fasta format in which the MiBiG sequences were used to construct reference databases and antiSMASH’s as query sequences for BLASTp. With the TSV files from BLASTp, we grouped all possible matches by each query sequence and selected the one with the lowest e-value. Finally, we used BioPython to plot the comparisons with more than 70% of similarity.

The BRA006 isolates were cultivated in 100 mL of the sterile A1 culture medium (Starch 10 g/L, Yeast extract 4 g/L, Peptone 2 g/L and Sea Water 75%), in 250 mL Erlenmeyer flasks. After 5–7 days under 120 rpm agitation and 28°C, the liquid cultures were extracted with ethyl acetate (1:1) under agitation for 2 h, and the organic phase was dried under pressure and kept at 4°C. For the LC-MS/MS analyses, the organic extracts were diluted in methanol, and the extract solution to be injected was prepared in methanol:water solution at a ratio of 1:1 (v/v) with a final concentration of 1.0 mg/mL (Bauermeister et al., 2016)⁠. The LC-MS/MS analysis itself was conducted in the Acquity UPLC H-Class (Waters, Milford, MA - US) hyphenated with Impact II mass spectrometer (Bruker Daltonics, Billerica - US). The mobile phase (flow 0.3 mL.min-1) consisted of water (A) and methanol (B) in the following gradient: 0.0–15.0 min (5%–20% B, curve 6); 15.0–30.0 min (20%–95% B, curve 6); 30.0–33.0 min - (100 B, curve 1); 33.0–40.0 min (5% B, curve 1). C18 - Luna (Phenomonex® - 100 mm × 2.1 mm × 2.6 μm) and the temperature adjusted to 35°C. The parameters adjusted for the spectrometer were: end plate offset of 500V; capillary voltage of 4.5kV; nitrogen (N₂) was used as gas; drying gas flow at 5.0 L.min⁻¹; drying gas temperature at 180°C; 4 bar nebulizer gas pressure; positive ESI mode. Spectra (m/z 30–2000) were recorded at a rate of 8 Hz. The quadrupole ion energy was set to 5.0 eV. The collision cell was set to 5.0 eV, with collision energies ranging from 20 to 50 eV (transfer time from 20 to 70 μs), and absolute fragmentation cutoff of 1,000 counts. Ions below 200 Da were excluded from fragmentation. The “active exclusion” function was enabled and with the following settings: exclude after 3 spectra; release after 0.3 min; reconsider precursor if the ratio current intensity to previous intensity was 1.8 (Bazzano et al., 2024). Accurate masses were obtained using sodium formate solution (10 mM) as a calibrant (Hoang et al., 2024).

The .csv and .mgf files generated in MZmine3 (Schmid et al., 2023) from the raw data of the metabolomic analysis were imported into the GNPS2 platform (https://gnps2.org/), where the molecular network and spectral pairing were performed using the Feature Based Molecular Network (FBMN) (Nothias et al., 2020) (https://gnps2.org/workflowinput?workflowname=feature_based_molecular_networking_workflow). We used the standard parameters for FBMN. Once the network was built, the annotations were propagated using the ChemWalker (Borelli et al., 2023) tool through GNPS2 interface (https://gnps2.org/workflowinput?workflowname=chemwalker_nextflow_workflow). We used the standard parameters for ChemWalker, including COCONUT (Sorokina et al., 2021) as the reference database and 0 for the component index to propagate information for the whole network. For the nodes that could not be annotated from these two previous methods, the MS/MS mass spectra of those compounds were analyzed using the SIRIUS tool version 5.8.5. To use this tool, we followed the annotation recommendations for Q-TOF mass analyzers and adopt the following adducts [M + H]⁺, [M + K]⁺, and [M + Na]⁺ as well as all the formulas available through the tool access (Dührkop et al., 2019). For in silico annotation, both ChemWalker and SIRIUS were used to annotate a structure with the best ranked candidate.

The isolate BRA006 exhibits very characteristic growth. The colonies are orange in color with apical growth and a rough appearance with the presence of individual dark-colored spores. In solid medium, it releases as yet undetermined compounds that diffuse into the agar, giving it a pink-purple color, as can be seen in Supplementary Figure 1.

We compared Illumina and MinION whole genome sequence techniques results from BRA006. While Illumina assembly yielded 10 contigs and 6,734,372 base pairs (6.73 Mb) in total, sequencing by MinION showed higher genome completeness, with an assembly resulting in 4 contigs with 6,762,267 base pairs (6.76 Mb) in total. Functional annotation performed by Prokka showed a significant increase in CDS predicted from MinION assembly data driven by a high amount of hypothetical proteins. We corrected this issue adding modiciations in MinION assembly pipeline and obainted similar values to Illumina. The complete result is shown in (Table 1). The CDS prediction followed the number of hypothetical proteins, with more predictions from MinION than from Illumina.

Micromonospora is one of the largest genera of Actinomycetota and possesses a large repertoire of bioactive secondary metabolites (SM) with a broad spectrum of therapeutic effects (Yan et al., 2022), for instance, aminoglycosides and macrolactam antibiotics. Through antiSMASH, we annotated the BGC content from Illumina and MinION assemblies of BRA006. MinION assembly resulted in a total of 15 BGCs (Table 2) that vary in similarity with antiSMASH database. Among them, there are those with reported antimicrobial, antifungal, and antitumor activities. For instance, quinolidomicin A is a macrolide with antibiotic and anticancer effects isolated from Micromonospora sp JY16 (Ruzzini and Clardy, 2016) and BRA006 presents a quinolidomicin BGC with 219 Mb in length and 67% similarity with the most similar known cluster.

The type-III polyketide Loseolamycin was identified from Micromonospora endolithica and inhibited the growth of the Gram-positive Bacillus subtilis and also showed herbicidal activity against the weed Agrostis stolonifera (Lasch et al., 2020). Cinerubins are anthracycline antimicrobials produced by actinomycetota which also present antitumor activity (Paderog et al., 2020; Silva et al., 2020). BRA006 possesses highly similar clusters to cinerubin B (74% by MinION and 80% by Illumina) and loseolamycin (88% by MinION and 92% by Illumina) A1/A2. BRA006 also has less similar BGCs for the biosynthesis of other natural compounds with antitumoral activity such as bleomycins and kedarcidin, isolated from Streptomyces verticillus and an unclassified Actinomycetales strain (ATCC 53650) Actinomycetes, respectively (Hecht, 2000; Hofstead et al., 1992).

AntiSMASH results from Illumina assembly yielded 18 BGCs (Table 2) in total, including clusters for the production of quinolidomicyn, cinerubin B, and loseolamycin A1/A2 found in both Illumina and MinION data. However, only by sequencing with Illumina, it was possible to find a BGC 100% similar to the production of sungeidines (Low et al., 2020), a group of metabolites produced by pathways with close evolutionary relations with the antitumor dynemicins (Unno et al., 1997).

Since both genome sequencing methods yielded identical clusters with enzymes from pathways for synthesizing compounds with medical applications, we decided to explore the evolutionary relationship. According to digital DNA:DNA hybridization used by Type Genome Server (TYGS) (Meier-Kolthoff and Göker, 2019) for phylogenetic inferencing, these two assemblies were classified as Micromonospora sp and pointed out as possible novel species due to their relatively high genomic distance to its closest related group: Micromonospora aurantiaca ATCC 27029 (Figure 2). However, the BGC with higher similarity to the antiSMASH database is the one for the production of loseolamycin A1/A2 in both MinION and Illumina assemblies. Therefore we used BLASTp to compare the proteins within those BGCs to the reference: BGC0002362 from Micromonospora endolithica.

Figure 2B shows the protein similarity between loseolamycin-producing BGC from BRA006 and M. endolithica. Both assembly methods detected the complete inversion of this M. endolithica BGC followed by a series of indels in the upstream region majorly composed of CDSs that encode proteins with no functional annotation. Other BGCs with less similarity to the antiSMASH database were also compared with the reference sequences. For the cinebubin B and quinalidomicin A BGCs there were differences between antiSMASH results regarding the sequencing method. In the case of cinerubin B BGC, sequencing by Illumina resulted in a BGC with 172 Mb length while MinION’s with 71 Mb. Both with a few highly similar proteins to the reference BGS (BGC0000212) (Supplementary Figure 2) original from Streptomyces sp. SPBO074. On the other hand, quinolidomicin A BGCs showed 219 Mb in MinION and 70 Mb in Illumina assembly. Quinolidomicin A BGC from MinION data presented more similarity (67% against 45% from the counterpart) to reference BGC0002520 original from Micromonospora sp (Supplementary Figure 3) although with shorter CDSs.

From the analysis of potential BGCs found in BRA006, where their potential to biosynthesize compounds with antibacterial (loseolamycins A, quinolidomicin A, cinerubin B, and brasilicardin A), antifungal (pradimicin A) and anticancer (bleomycin A2 and kedarcidin) activity was identified, we investigated whether these compounds were present in the metabolome of this Actinomycetota. To do so, and to extend the analysis to other compounds that BRA006 might be able to produce, we performed a metabolomics analysis in which the BRA006 extract was analyzed by LC-ESI(+)-HRMS/MS. After the LC-MS/MS analysis, the obtained raw data were converted into. mzXML format using Proteowizard (Chambers et al., 2012), and feature finding was performed using MZmine (Schmid et al., 2023). We then performed the annotations sequentially, in three steps. In the first, the annotations were made by spectral pairing and molecular network construction through GNPS2 (https://gnps2.org/). As a result, we obtained 527 nodes with at least one connection, of which 83 were annotated by GNPS2. Based on this result, we propagated the annotations to the nodes that did not present an annotation by spectral pairing using ChemWalker (Borelli et al., 2023), increasing the annotations to 373 nodes. Finally, for the nodes that could not be annotated by this tool, we used a third in silico spectral annotation tool, SIRIUS (Dührkop et al., 2019), increasing the annotations for an additional 67 nodes. Thus, a total of 527 nodes were represented by the molecular network, of which 523 were annotated. With these results in hand, we performed an automated chemical classification of each annotated compound using ClassyFire (Djoumbou Feunang et al., 2016). The molecular network was colored based on the superclasses to which each annotated compound belonged (Figure 3).

The compounds could be grouped into 13 different chemical superclasses, where 30.2% of the annotated compounds belong to the superclass of lipids and lipid-like molecules, 19.5% to organoheterocyclic compounds, 15.1% to organic acids and derivatives, 9.9% to benzenoids, 7.6% to phenylpropanoids and polyketides, 6.5% to organic oxygen compounds, 4.0% to organic nitrogen compounds, 1.7% to alkaloids and derivatives, 0.6% to lignans, neolignans, and related compounds, 0.4% to hydrocarbon derivatives, 0.2% to organic 1, 3-dipolar compounds, 0.2% to hydrocarbons, 0.2% to organosulfur compounds, and 3.8% could not be classified (None). The complete relationship between all annotated compounds and their chemical hierarchical classification into kingdom, superclass, and class is shown in Supplementary Table 4. From the entire set of annotations and classifications, we searched for the seven compounds predicted by antiSMASH in the BRA006 metabolome. None of the seven compounds were found in the annotation pool. Therefore, we took the molecular structure of these seven compounds and performed their chemical classification using ClassyFire; once we had their chemical class, we would search for compounds annotated in the BRA006 metabolome that had the same chemical class. From a pharmacological point of view, compounds belonging to the same chemical class might belong to the same biosynthetic pathway and or have similar effects, as is the case, for example, with the class of peptidomimetics in the treatment of cancer (de Valk et al., 2020) and steroids in the treatment of pain (Paulsen et al., 2013). The seven compounds were organized into six different classes of molecules, where loseolamycin A1 belongs to the phenol class, cinerubin B belongs to the anthracycline class, brasilicardin A belongs to the steroid and steroid derivative class, pradimicin A belongs to the naphthacene class, bleomycin A2 belongs to the peptidomimetic class, and quinolidomycin A and kedarcidin are organooxygen compounds.

Of these six chemical classes to which the compounds of interest belong, the Phenols class presented 10 annotated molecules, while the Organooxygen Compounds and Steroids and Steroid Derivatives classes presented 34 compounds each and Peptidomimetics presented two compounds in our analysis. Among the 10 compounds belonging to the Phenol class, three of them had bioactivity previously reported, but with a different action from the antibiotic loseolamycins A1. The targeting offered by antiSMASH, by searching for molecules belonging to the same class as those with bioactive activity predicted by the tool, allowed us to find a larger and more diverse range of compounds.

Using a reverse flow of integrative analysis, where we start from what was annotated in the metabolome, we set out to evaluate whether it would be possible to identify, from a given metabolite, the enzymes that lead to its production in BRA006. To this end, we first crossed the annotated metabolome with the KEGG database (Kanehisa and Goto, 2000) to search within the metabolome of this Actinomycetota for molecules with known biosynthetic pathways. As a result, we found a molecule belonging to the staurosporin biosynthetic pathway, brevianamide F. It should be noted that of the three methods used in the annotation process, brevianamide F was annotated by spectral pairing with the GNPS spectral library (Wang et al., 2016), showing a MQScore of 0.97 and a m/z error of 2.47ppm. Since the topology of the molecular network is given by the similarity between nodes, the node with an annotation for brevianamide F is connected to seven other nodes (Figure 3). All seven nodes have been annotated by ChemWalker. Looking at the predicted structures, six of the seven share the same indole-like nucleus as brevianamide F, suggesting that other molecules may ultimately be produced, either by brevianamide F BGC or by other intermediates belonging to the same pathway as brevianamide F. Once identified, the enzymes that make up the staurosporin biosynthetic pathway, we searched the BRA006 genome for which of these enzymes would be encoded. From this search, 16 staurosporin pathway enzymes were identified in the genome of this Actinomycetota (Supplementary Figure 4), and 11 overlap the region of the BGC 20 from Illumina (Figure 4). Among them, we found an NRPK 2,3-dihydroxybenzoate-AMP ligase functionally classified as a biosynthetic-additional enzyme by antiSMASH. These results show that a dynamic integrative approach, i.e., first combining spectral and in silico annotation, assigning chemical classes, and then searching for metabolites from the genome to match encoded proteins in the genome to the pathway producing the putative metabolite annotated, is an efficient approach to characterize new species with the potential to produce bioactive compounds.