Genome mining and UHPLC-MS/MS illuminate the specicity of secondary metabolite synthetic gene clusters in Bacillus subtilis NCD-2

Bacillus subtilisstrain NCD-2 is anexcellent biocontrol agent against plant soil-borne diseases and shows broad-spectrum antifungal activities. This study aimed to explore all the secondary metabolite synthetic gene clusters and related bioactive compounds in NCD-2. An integrative approach, which coupled genome mining with structural identication technologies using ultra-high-performance liquid chromatography coupled to quadrupole time-of-ight tandem mass spectrometry (UHPLC-MS/MS), was conducted to interpret the chemical origins of the signicant biological activities in NCD-2. secondary


Background
Bacillussubtilisand its closely related speciesare ubiquitousinhabitantsof soil,andare widely recognized as powerful biocontrol agentsagainst plant soil-borne diseases [1]. The Bacillusgenus has received considerable attention as a biological resourceused in the development of microbial pesticides, in part because its members form stress-resistant spores that do not harm the environment and are useful in pesticide production [2][3][4]. The mechanisms used by B. subtilisto suppress plant soil-borne diseases include competing with phytopathogens for nutrients and spatial sites, inducing the systematic resistance of plants,and inhibiting pathogen growth by producing antimicrobial compounds [5]. The latter is a general characteristicof B. subtilis' biocontrol capability and plays an important role in the biological control of plant diseases [6,7]. B. subtilis produces more than two dozen antimicrobial compounds having amazing structural variety. On the basis of the biosynthetic pathway, the antimicrobial compounds are divided into small molecular compounds synthesized by the ribosomal pathway, such as bacteriocins, and peptide compounds synthesized by the non-ribosomal pathway, such as lipopeptidesand polyketones [8]. Most antimicrobial compounds are secondary metabolites,with very complex chemical structures, that are not necessary for the growth and reproduction of microorganisms.
Secondary metabolites function as essential chemical signals for the induction of cellular differentiation in the producing organism and for controlling its metabolism [9,10]. They also function as antibiotics,and their antimicrobial properties may lead to shifts within rhizospheric microbial functional subsystems, such as affecting the availability of nutrients for the plant [11].
The genes encoding the secondary metabolites commonly exist in clusters and encode complex enzymes with multiple functions [12]. Thepolyketide synthase/non-ribosomal peptide synthase (PKS/NRPS) gene clusters have been well studied. The PKS pathway peptides require at least three domains, an acyl transferase, a ketosynthase, and an acyl carrier protein [13]. The NRPS pathway shares a common mode of synthesis, the multicarrier thiotemplate mechanism,requiring the cooperation of three basic domains [14]. The adenylation domain selects its cognate amino acid and generates an enzymatically stabilized aminoacyl adenylate. The peptidyl carrier domain is equipped with a 4′-phosphopantetheine prosthetic group to which the adenylated amino acid substrate is transferred and bonded by a thioester bond. The condensation domain catalyzes the formation of a new peptide bond [13].The carbon skeleton in the metabolite is synthesized by the core PKS and NRPS enzymes, and then, the nal product is formed with the assistance of various modifying enzymes [15]. The bioactive secondary metabolites produced by the PKS/NRPS pathway in species of B. subtilisinclude bacilysin [16], bacilysocin [17], surfactin [18], iturin A [19], fengycin [20], mycosubtilin [21], bacillomycins [8], and di cidin [16].
The traditional method of screening for new active products is based on testing for biological activity.
However, this method is time-consuming and the same products have been repeatedly discovered [22]. Thus, the discovery of natural products had encountered a bottleneck [23], and the development of a more rapid and effective screening strategy to detect new secondary metabolites was necessary [24,25].
Genome mining is a technology that uses modern bioinformatics to recognize speci c functional genes or gene clusters from genome sequences [26]. With the rapid development of gene sequencing technology and the decreasing cost of genome sequencing, increasing numbers of microbial genome sequences have been determined [27]. Therefore, genome mining has become a more accurate and e cient screening strategy for discovering new metabolites [26].
B. subtilis strain NCD-2 is a promising biological control agent against plant soil-borne diseases that produces lipopeptides,fengycin, and surfactin [28]. Fengycin has an antifungal activity, and surfactin facilitates the root colonization ability of strain NCD-2. Both fengycin and surfactin play important roles in strain NCD-2's ability to suppress plant soil-borne diseases [29]. The purpose of this study was to identify potential secondary metabolites with antifungal activities in strain NCD-2 using genome mining. Then,a bioinformatics analysis was conducted to reveal the differences between gene clustersfor these secondary metabolites in strain NCD-2 and reference strainBacillusvelezensisFZB42. Finally, ultra-highperformance liquid chromatography coupled to quadrupole time-of-ight tandem mass spectrometry (UHPLC-QTOF-MS/MS) was used to identify the different homologs of fengycin and surfactinfound in strain NCD-2.

Results
Genomic features of strain NCD-2 A total of 501,671,500 paired-end reads and 5,016,715 clean single reads (412-bp library; paired-ends of 75 bp) were assembled using the software Velvet [30]. The genome of B. subtilis NCD-2 contained 189 contigs (> 133 bp; N90, 16,187) of 4,644,322 bp, with an average G + C content of 63.74%. The nal assembled genome comprised 4,444 genes, including 4,329 protein-coding genes (418 signal peptidecoding genes), 83 tRNA genes for all 20 amino acids, 30 rRNA genes, and 2 CRISPR repeat genes. A total of nine putative gene clusters responsible for antimicrobial metabolite biosynthesis were identi ed. These gene clusters included PKS and NRPSgenes (Fig. 1).
The taxonomic status of strain NCD-2 At present, 272 B. subtilis genome sequences are deposited in the GenBank database, including 113 whole-and 159 incomplete genome sequences. The genome sizes of the 272 B. subtilis strains range from 2.68 Mb to 5.35 Mb, and the GC contents range from 42.9-46.6%. These genome sequences were downloaded from the GenBank database, and their accession numbers are listed (Additional le 1, Table   S1). To analyze the evolution of different B. subtilis strains, a phylogenetic tree was constructed based on the genome sequences. The 272 strains of B. subtilis were divided into four subspecies, subtilis, inaquosorum, spizizenii, and stercoris [31]. As shown in Fig. 2, strain NCD-2 (represented by the black bar) clustered together with B. subtilis strain UD1022 and was closely related to B. subtilis strains XF-1, BAB-1, HJ5, SX01705, and BSD-2.
Secondary metabolite biosynthetic gene clusters in strain NCD-2 The secondary metabolite biosynthetic gene clusters in the genome of strain NCD-2 were predicted using the online website antiSMASH [32]. In total, nine secondary metabolic gene clusters were identi ed in the NCD-2 genome sequences (Table 1), including three NRPS, two terpenes, one heterozygous Nrps-Transatpks-Otherks, one type III polyketide gene cluster, one Sactipeptide-head to tail, and a gene cluster with an unknown function. The structural compositions of the gene clusters are shown in Fig. 3. These clusters were composed of core biosynthetic, additional biosynthetic, transport-related, regulatory, and other genes. Among these nine gene clusters, clusters 3,7, 8, and 9 had 100% amino acid sequence homology with known gene clusters that synthesize bacillaene, bacillibactin, subtilosin, and bacilysin, respectively (Table 1). Gene cluster 1 showed 82% amino acid similarity with a surfactin synthetase gene cluster, and gene cluster 4 showed 93% amino acid similarity with a fengycin synthetic gene cluster in B. velezensis strain FZB42. However, gene clusters 2, 5, and 6 did not match any known gene clusters. Clusters 1 and 4 of strain NCD-2 were further compared with those of the model strain 168 and B. subtilis strains closely related phylogenetically to strain NCD-2. The fengycin synthetic gene cluster of strain NCD-2 contained three genes, fenEAB, while the other strains contained ve genes, fenCDEAB (Additional le 1, Fig. S1). SrfAB of surfactin was synthesized by the normal transcription and translation of srfAB in the 11 strains. However, the same SrfAB was assembled with Gms0365 and Gms0366 and then transcribed and translated by gms0365 and gms0366 separately in strain NCD-2 (Additional le 1, Fig.  S2). Therefore, we hypothesized that the structures and functions of fengycin and surfactin from strain NCD-2 may be different from those of the other B. subtilis strains. Speci city of surfactin and fengycin synthetase gene clusters in B. subtilis NCD-2 The surfactin synthetic gene cluster in strain NCD-2 was analyzed using PRISM, and the core genes were selected for a PKS/NRPS analysis. This gene cluster contained four genes: gms0365, gms0366, gms0367, and gms0368. Gms0365 showed an identical conserved structural and functional domain, CATCATCATe, with SrfAA in strain FZB42, in which C, A, T, and Te represent the condensation, adenylation, thiolation, and thioesterase domains, respectively (Fig. 4a). Compared with SrfAB in strain FZB42, Gms0366 in strain NCD-2 had lost the T and E domains, but the amino acid residues for the binding pockets of Gms0366 were exactly the same as those of SrfAB. The residues of the different adenylation domains A6 and A2 from the enzymes Gms0365 and Gms0366, respectively, were exactly the same, and both bound the amino acid leucine. Gms0367 had only T and E domains, with no speci c substrate-binding domain. The superposition of Gms0367 and Gms0366 domains formed a complete SrfAB.The T domain was reversed between Gms0367 and Gms0368. The domains of Gms0368 were CATe, in which the thioesterase domain releases linear peptide chains. The domains of Gms0368 were exactly the same as those of SrfAC, but the amino acid residues forming the binding pockets were not completely conserved. The residue sequence was DAF-LGCV, compared with DAFXLGCV of strain FZB42, revealing a difference of one residue.
The fengycin synthetic gene cluster was analyzed by PRISM, and the core genes were selected for a PKS/NRPS analysis. This cluster contained ve genes in strain FZB42's genome, they were ordered as fenCDEAB (Fig. 4b). However, the fengycin synthetic gene cluster in strain NCD-2 contained only three genes: gms1961, gms1960, and gms1958. Gms1961 of strain NCD-2 corresponded to FenE in strain FZB42 and they had conserved residues of A8, which bound two amino acids. Gms1960 and Gms1959 in strain NCD-2 had amino acids sequences identical to FenA and FenB in strain FZB42, respectively. Interestingly, no homologs of FenC and FenD were identi ed in the genome of strain NCD-2. Consequently, the amino acid sequences ofFenC and FenD from strain FZB42 were compared with the strain NCD-2 proteome using BioEdit. Gms1961 was most similar to FenC, and Gms1960 was most similar to FenD(Additional le 1, Tables. S2, S3). Therefore, it was hypothesized that Gms1961 and To further investigate whether the structure of the fengycin synthetase gene cluster in NCD-2 is strain speci c, the fengycin synthetic gene clusters from 11 different B. subtilis strains that are closely related to strain NCD-2 or are model strains were compared (Additional le 1, Fig. S1). The gene cluster sequences of all 11 strains were ppsABCDE (also fenCDEAB), and only that of strain NCD-2 was fenEAB. Therefore, the fengycin synthetic gene cluster of strain NCD-2 is unique.
Fengycin was separated from the lipopeptide extract of strain NCD-2 using Fast protein liquid chromatography (FPLC)(Additional le 1, Fig. S3), and the QTOF-MS/MS analysis revealed ve fractions in the fengycin cluster ( Fig. 5a- ions (α and β), representing the linear N-terminal and the cyclic C-terminal segments, respectively, of diverse fengycin species (Additional le 1, Fig. S4AB) and (Fig. 5a-e). The MS/MS spectrum of the fengycin ion at m/z 732.4 yielded two intense product ions at m/z 966.5 and 1,080.5, representing fengycin A (Fig. 5a), while the MS/MS spectrum of the fengycin ion at m/z 746.4 (Fig. 5b) yielded key product ions at m/z 994.5 and 1,108.6, representing fengycin B (Fig. 5b). The MS/MS spectrum of the fengycin ion at m/z 725.4 yielded two intense product ions at m/z 952.4 and 1,066.5, representing fengycin A2 (Fig. 5c), while the MS/MS spectrum of the fengycin ion at m/z 739.4 (Fig. 5d) yielded key product ions at m/z 980.5 and 1,094.5 representing fengycin B2 (Fig. 5d). The MS/MS spectrum of the fengycin ion at m/z 767.4 yielding two intense product ions at m/z 994.5/1,008.5 and 1,108.6/1,122.6 representing fengycin C (Fig. 5e). Five classes of fengycins were identi ed based on the key product ions of β-hydroxy fatty acid (β-OH FA) with chain lengths varying from C12 to C20 (  Fig. S4C) and (Fig. 5f). Based on these key product ion, one class of surfactin was identi ed, which were the surfactins (m/z values of 994.6, 1,008.7, 1,022.7, and 1,036.7) of fatty acids with chain lengths varying from C11 to C15 (Fig. S10).

Discussion
Species of B. subtilis have the potential to produce two dozen antimicrobial substances, and 5-8% of the B. subtilis genome contributes to the production of antimicrobial substances [33]. Some inhibit the growth of pathogens and the germination of spores. The lipopeptide mixture of B. subtilis C232 inhibits the formation of Verticillium dahliae microsclerotia [34], and the volatile compounds secreted by B. subtilis JA inhibit the conidial formation and mycelial growth of Glomus etunicatum [35].
However, certain bioactive compounds are synthesized only under special conditions or as the result of external stimulation; therefore, it is di cult to obtain all the antimicrobial compounds produced by Bacillus using traditional cultivation and extraction methods, and this limited the comprehensive understanding of the mechanisms of biological control and biocontrol bacteria [22]. Genome mining allows the prediction of metabolites based on genome sequences and is widely used in obtaining new antibiotics [26]. It was used to identify a new NRPS pathway product, coelichelin, in Streptomyces coelicolor [36]. Pseudomycoicidin in Bacillus pseudomycoides DSM 12442 was discovered through the heterologous expression of its BGC in Escherichia coli [37]. Traditional cultivation and extraction methods were used to identify lipopeptide, fengycin, and surfactin from B. subtilis NCD-2, and fengycin showed strong antifungal abilities against V. dahliae and B. cinerea. However, the fengycin-de cient mutant of strain NCD-2 still has a certain antifungal ability, but it is less than that of wild-type strain NCD-2.
Therefore, other antifungal active compounds, besides fengycin, may be produced by strain NCD-2 [2,28,29]. In this study, genome mining was conducted to analyze the potential antimicrobial compounds of the NCD-2 strain, and some of them were identi ed using MS. In total, nine kinds of secondary metabolite gene clusters related to surfactin, bacillaene, fengycin, bacillibactin, subtilosin, bacilysin, two terpenes, and one unknown product were identi ed from the genome of strain NCD-2. The surfactin [38], bacillaene [39], fengycin [40], bacilliactin [41], subtilosin [42], and bacilysin [43] showed antimicrobial abilities and played different roles in suppressing plant diseases. Only fengycin and surfactin were identi ed from the lipopeptide extract of NCD-2 despite the presence of other gene clusters in the genome. These other antimicrobial compounds may not have been detected because the acid precipitation extraction method was not suitable. Some bioactive compounds, such as bacillaene, bacillibactin, subtilosin, and bacilysin, are not lipopeptides. Therefore, these substances were not extracted using hydrochloric acid precipitation [44][45][46][47].
Fengycin comprises a peptide ring circled by 10 amino acids with a fatty acid chain tail. The mechanism of fengycin synthesis has been well studied in B. velezensis strain FZB42 [48], and the fengycin synthetic gene cluster in the strain consists of ve genes (38 kb) that encode the synthetases FenCDEAB, of which FenC recognizes and carries glutamate and ornithine, FenD recognizes and carries tyrosine and threonine, FenE recognizes and carries glutamate and valine, FenA recognizes and carries proline, glutamine, and tyrosine, and FenB recognizes and carries isoleucine. FenCDEAB recognizes 10 amino acids and carries them to the β-OH FA chain to form fengycin [49][50][51]. However, NCD-2 only had fenEAB, lacking fenC and fenD, compared with the typical cluster structure of fenCDEAB in the FZB42 strain and 10 other Bacillus strains (Fig. 4b) and (Additional File 1,Fig. S1). To identify the enzymes FenC and FenD in the NCD-2 genome, their amino acid sequences from FZB42 were selected to screen for homologs by scanning the local NCD-2 proteome using BioEdit. The Gms1961 protein in the NCD-2 strain had the greatest similarity to FenC at an amino acid sequence level (Additional File 1, Table S2). The Gms1961 protein contained 2,550 amino acids, and the molecular weight was 287.50 kDa. The substrate bound by the adenylation domain of the Gms1961 protein was predicted (Additional File 1, Table S4). The adenylation A9 domain bound valine and N5-hydroxyornithine, with the latter being a transitional form of ornithine combined with the adenylation domain [52]. The UHPLC-QTOF MS/MS of the fengycins revealed that all the structures possessed the amino acid ornithine at position 2 ( Fig. 5a-e), indicating that there was a protein that transports ornithine in the NCD-2 strain. Thus, it was hypothesized that Gms1961 functions as FenC and FenE. The analysis was performed using the Gms1960 protein and it had the greatest similarity with FenD (Additional File 1, Table S3); however, the FenD domains in Gms1960 and FZB42 varied greatly. Therefore, it was hypothesized that Gms1960 or other enzymes may have functions similar to those of FenD.
Although the fengycin synthetic gene cluster in the NCD-2 strain lacked two important genes-fenC and fenD-hat synthesize enzymes compared with the reported fengycin synthetic gene cluster, the NCD-2 strain was capable of producing 26 homologs of 5 kinds of fengycins. The amino acids at position 6 and 10 of the fengycin cyclic peptide ring determine the type of fengycin. There are currently ve types of reported fengycins, A, B, A2, B2, and C (Additional File 1, Fig. S4). When the amino acid at position 6 was valine and at position 10 was isoleucine or valine, then fengycin B or fengycin B2, respectively, was produced ( Fig. 5a, b) and (Additional File 1, Fig. S4); however, if the amino acid at position 6 was alanine, then fengycin A or fengycin A2, respectively, was produced (Fig. 5c, d) and (Additional File 1, Fig. S4). When the amino acid at position 6 was isoleucine or leucine and at position 10 was valine, then fengycin C was produced (Fig. 5e) and (Additional File 1, Fig. S4). The MS analysis of the fengycins in the NCD-2 strain revealed that the strain was capable of producing these ve kinds of fengycins. Based on differences in the number of carbon atoms in the β-OH FA, fengycin had different homologs, and the molecular weight of each homologs differed by 14 (-CH2) [53]. The molecular structure of the lipopeptide determines its biological activity, and long-chain fatty acids increase the hydrophobic activities of lipopeptides, making them more likely to have membrane-bound antimicrobial effects [54]. A Bacillus circulans strain produces four fengycin homologs, but only fengycins with C16 and C17 carbon atoms in their β-OH FA chains had antibacterial activities [55]. The NCD-2 strain produced 14 fengycin homologs having more than 16 carbon atoms, and they accounted for a large proportion of all the homologs. It was speculated that these long-chain fengycins play important roles in the antimicrobial functions of NCD-2.
The Bacillus siamensis SCSIO 05746 strain produces a great number of fengycin homologs, including 19 homologs of fengycin B [56]. Using an MS analysis, the ve fengycins produced by the NCD-2 strain were divided into 26 homologs (Fig. 5a-e) and (Additional File 1, Fig. S5-S9). Therefore, NCD-2 is currently the strain with the largest number of known fengycin homologs [57].
During the microbial synthesis of secondary metabolites, such as lipopeptide, the relatively high energyconsuming process of protein synthesis takes priority [58]. Excessive energy consumption is not conducive to the normal growth of microbes, and, generally, microbes produce antibiotics in large amounts only when encountering pathogens or other stresses [59]. In the long-term evolution of NCD-2, the key synthetic genes fenEAB involved in synthesizing fengycin were conserved, while two important synthetic genes fenCD were lost. However, ve fengycins are still produced. Gms 1961 played the dual roles of FenC and FenE, indicating that NCD-2's fengycin synthetic process, which is unique to the strain, was more energy-e cient than the process used in the other strains..

Conclusions
In this study, genome mining and UHPLC-QTOF-MS/MS were performed. They determined that there were more gene clusters encoding antimicrobial compounds in the genome of the NCD-2 strain and that the fengycin synthetic gene cluster was unique. The results indicated that the NCD-2 strain has a unique mechanism for synthesizing fengycin. Using molecular genetics and biochemistry to analyze the new mechanism of fengycin synthesis may provide a new theory for the synthesis of antimicrobial compounds through the NRPS pathway.

Microorganisms and culture conditions
Bacillus subtilis NCD-2 was routinely grown at 37 °Con Luria Bertanimedium. For secondarymetabolite production, strain NCD-2 was grown in Landy brothat 30℃ and180 rpm [60]. Phytopathogen Botrytis cinereaBC-10 was used for antifungal activity testsfollowing the method described by Guo [29] with some modi cations. Brie y, a 6-mm diameter disc of B. cinerea was placed in the center of a 9-cm potato dextrose agar (PDA) plate, and the plates were inoculated with B. subtilis NCD-2 using a sterilized toothpick 2 cm from the center. Finally, the diameter of the inhibition zone was measured after a 3-d incubation at 25℃.
Predictions and a speci city analysis of secondary metabolite synthetic gene clusters Secondary metabolite synthetic gene clusters for strain NCD-2 were detected using antiSMASH (http://antismash.secondarymetabolites.org) [32,72] and PRISM (http://grid.adapsyn.com/prism/) [73] with the parameters selected by default. Functional domain predictionsfor PKS/NRPS in the predicted gene clusters were analyzed using the PKS/NRPS Analysis Website (http://nrps.igs.umaryland.edu/) [74]. Typical PKS and NRPS sequences were selected for genomic and proteomic scanning after using BioEdit software to create a local BLAST based on strain NCD-2's genome and proteome, respectively.

Separation of lipopeptides by FPLC
Lipopeptideswere extracted using the method described by Guo [29]. Brie y, strain NCD-2 was culturedin Landy broth [60]at 30℃ for 72 h with shaking at180 rpm. The cell-free supernatant was obtained by centrifugation at 8,000 × g for 30 min at 4℃. The supernatant was adjusted to pH 2.0 with 6 mol/L HCl and stored for 12 h at 4℃. After centrifugation at 10,000 × g, for 20 min, the resulting pellet was extracted with methanol under continuous magnetic stirring for 2 h. The obtained extracts were sterilized by passing through 0.45-µm lters (Millex-GV, Millipore, Billerica, MA, USA) to obtain crude lipopeptides.The crude lipopeptides were separated and puri ed using an AKTA Puri er(GE Healthcare, Uppsala, Sweden) with the SOURCE 5RPC ST 4.6/150 column as described previously [75]. The lipopeptides were eluted by solvent A [2% acetonitrile containing 0.065% tri uoroacetic acid (TFA) (V/V)] and solvent B [80% acetonitrile containing 0.05% TFA (V/V)] using a linear gradient of 0-100% acetonitrile over 57 min at a ow rate of 1 mL/min. The detection wavelength was 215 nm. All the main peaks were collected by FPLC automatically. Finally, each peak was concentrated using a rotary evaporator and was analyzed using UHPLC-QTOF-MS/MS.
The MS analysis was performed using a 5600 TripleTOF system equipped with a DuoSpray™ Ion Source, and the data were processed using Analyst TF 1.7 software (Applied Biosystems Sciex, Toronto, ON, Canada). PeakView™ software 2.0 (Applied Biosystems Sciex, Toronto, ON, Canada) was used for investigating and interpreting mass spectral data with special tools for processing accurate mass data and structural elucidation. The DuoSpray™ ion source was used in positive ion mode.The instrumental parameters were set as follows: ion spray voltage oating, 5,000 V; nebulizing gas, 50 psi; heater gas, 50 psi; curtain gas, 35 psi; temperature, 350℃; declustering potential (in TOF MS Table S1. All the B. subtilis strain with the assembly level of chromosome and their RefSeq assembly accession. Table S2. The homologues of FenC of FZB42 by scaning the local NCD-2 proteome in BioEdit.   Schematic diagram of nine secondary metabolite synthetic gene clusters in Bacillus subtilis strain NCD-2.
Different color blocks represent genes with different functions; the genes marked with dark red, light red, blue, green, and gray are core biosynthetic, additional biosynthetic, transport-related, regulatory, other genes, respectively.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.