Overview of the Antimicrobial Compounds Produced by Members of the Bacillus subtilis Group

Over the last seven decades, applications using members of the Bacillus subtilis group have emerged in both food processes and crop protection industries. Their ability to form survival endospores and the plethora of antimicrobial compounds they produce has generated an increased industrial interest as food preservatives, therapeutic agents and biopesticides. In the growing context of food biopreservation and biological crop protection, this review suggests a comprehensive way to visualize the antimicrobial spectrum described within the B. subtilis group, including volatile compounds. This classification distinguishes the bioactive metabolites based on their biosynthetic pathways and chemical nature: i.e., ribosomal peptides (RPs), volatile compounds, polyketides (PKs), non-ribosomal peptides (NRPs), and hybrids between PKs and NRPs. For each clade, the chemical structure, biosynthesis and antimicrobial activity are described and exemplified. This review aims at constituting a convenient and updated classification of antimicrobial metabolites from the B. subtilis group, whose complex phylogeny is prone to further development.


INTRODUCTION
The genus Bacillus comprises 377 species 1 (last update in January 2019) of Gram-positive, rod-shaped bacteria (Gordon et al., 1973). Their ability to form endospores, their diversity in physiological properties, as well as their capacity to produce numerous antimicrobial compounds (AMCs) favor their ubiquitous distribution in soil, aquatic environments, food and gut microbiota of arthropods and mammals (Nicholson, 2002).
Bacteria from the Bacillus subtilis group consist of small vegetative cells (<1 µm-wide) for which the strain B. subtilis subsp. subtilis 168 is considered as model organism (Barbe et al., 2009). They are usually mesophilic and neutrophilic, although some can tolerate high pH. The four original species of the group (B. subtilis, Bacillus licheniformis, Bacillus pumilus, and Bacillus amyloliquefaciens) were discovered more than 40 years ago (Gordon et al., 1973;Priest et al., 1987). Since then, the evolution of their molecular, chemotaxonomic and physiological characterizations led to regular reevaluations and (re-)description of numerous novel species and subspecies (see current taxonomy of the group in Figure 1) (Fan et al., 2017).
The potential of B. subtilis group strains to produce a wide diversity of secondary metabolites mediating antibiosis was recognized for decades. For any given strain of the B. subtilis group, it is now estimated that at least 4-5% of its genome is devoted to antimicrobial compounds (AMCs) production (Stein, 2005). These molecules are mainly antimicrobial peptides (AMPs). Their structures are usually cyclic, hydrophobic and contain peculiar moieties such as D-amino acids (AA) or intramolecular thioether bonds. In addition to AMPs, volatile metabolites also constitute a large family of antimicrobials exhibiting numerous metabolic and functional roles.
Due to the wide diversity of these molecules, their classification is rather complex and can be based on several criteria such as their biosynthetic machinery, sources, biological functions, properties, three-dimensional structure, covalent bonding pattern or molecular targets (Tagg et al., 1976;Wang et al., 2015). Here a classification of the B. subtilis group antimicrobial molecules is proposed, based on their biosynthetic pathways and their chemical nature as shown in Figure 2. This review will emphasize the biosynthesis pathway and the bioactivity of the main clades of AMCs within the B. subtilis group: i.e., the ribosomal peptides (RPs) (bacteriocins and enzymes), the polyketides (PKs), the non-ribosomal peptides (NRPs) and the volatiles. A full overview of this chart is provided as Supplementary Material (Supplementary Figure S1).

RIBOSOMAL PEPTIDES
Ribosomally synthesized peptides (RPs) are usually derived from short precursors (ca. 100 AA) and are processed to mature compounds through post-translational modifications (Oman and van der Donk, 2009). Various enzymes mediate these modifications and therefore generate a wide diversity of chemical structures. Most of these peptides were originally referred to as "bacteriocins, " characterized as low molecular weight molecules that exhibit inhibiting growth activities against bacteria closely related to the producing strain (Klaenhammer, 1988;Chopra et al., 2015). In addition to bacteriocins, other types of enzymes exhibiting antagonistic activities are also ribosomally synthesized. However, those compounds display diverse metabolic activities such as quorum sensing (QS) mediation, cell lysis or induction of genetic competence (Schmidt, 2010;Shafi et al., 2017). It should also be noted that molecules referred to as BLIS (bacteriocinslike inhibitory substances) include AMPs for which the ribosomal synthesis has not been confirmed yet (Abriouel et al., 2011).

B. subtilis Group Bacteriocins
It is estimated that 99% of the bacteria and archaea are able to produce at least one bacteriocin. Historically, lactic acid bacteria (LAB) were studied as main bacteriocin producers, mostly because of their long history of safe use in food fermentation (O'Sullivan et al., 2002). Nisin (Figure 3C), produced by Lactobacillus lactis subsp. lactis, was approved as a food additive in the 1960s and has since then been used in over 50 countries for its antimicrobial activity against Gram-positive pathogens such as Clostridium spp. and Bacillus spp. (Klaenhammer, 1988;Delves-Broughton, 1990). However, the search for new bioactive molecules has rapidly expanded to other bacteriocin-producing genera, with a particular attention, in the late 1990s, to the GRAS (generally recognized as safe) Bacillus species whose bacteriocin antimicrobial spectra were broader than those of LAB (Pedersen et al., 2002;Riley and Wertz, 2002;Sumi et al., 2015).
The generic biosynthetic pathway of Bacillus species bacteriocins includes several post-translational modifications, including the proteolytic cleavage of the leader peptide at the N-terminal end (McIntosh et al., 2009). The modifications of active peptides, its secretion and the immunity to the bacteriocin (as described below) vary depending on the bacteriocin class.
While many classifications have been suggested over the years, one reasonable way to cope with the diversity of the Bacillus bacteriocins is to sort them on the basis of their biosynthetic pathway as previously reported for Streptococcus spp. and Enterococcus spp. bacteriocins (Nes et al., 2007) and reviewed in Abriouel et al. (2011). Accordingly, three main classes subdivided into several subclasses can be distinguished for the B. subtilis group. As detailed in Table 1, Class I includes the posttranslationally modified peptides such as the lantibiotics whereas the non-modified peptides are grouped in Class II; Class III involved bacteriocins larger than 10 kDa (Abriouel et al., 2011). Supplementary Table S1 summarizes the different RPs produced by the strains belonging to the B. subtilis group, as well as their reported antimicrobial activities.
Class I includes small AMPs (19-38 AA) with extensive post-translational modifications. Subclasses I.1, I.2, and I.3 have in common their lantibiotic structure, which refers to interresidual thioester bonds made of modified AA residues. As illustrated in Figure 3, lantibiotics involve 2,3-didehydroalanine (Dha) and (Z)-2,3-didehydrobutyrine (Dhb), resulting from the dehydration of serine and threonine residues, respectively. The intra-molecular addition of Dha or Dhb on a cysteine residue leads to the respective formation of lanthionine and methyllanthionine bridges (Willey and Donk, 2007). Subtilin ( Figure 3B), from subclass I.1, is one of the most studied bacteriocins from the B. subtilis group. Its structure shares several similarities with nisin A lantibiotics, shown in Figure 3C (Guder et al., 2000;Abriouel et al., 2011). Peptides from subclass I.4 undergo other types of modifications. For instance, subtilosin A is a head-to-tail cyclic peptide with unusual inter-residue linkages (i.e., Cys-Phe bond) (Marx et al., 2001;Kawulka et al., 2004).
Class II bacteriocins include small (<10 kDa), linear and nonmodified peptides, resistant to heat and acido-basic treatments. They are divided in three subclasses based on a conserved AA motif near their N-terminus. The YGNGVXC (X is any AA) motif is associated to pediocin-like peptides from subclass II.1 whereas DWTXWSXL is specific to thuricin-like peptides from subclass II.2. Subclass II.3 comprises the small non-modified AMPs without any typical motif in their AA sequence (Abriouel et al., 2011). Finally, class III bacteriocins consist into large and heat labile molecules, generally characterized by a phospholipase activity .
Because of their wide diversity, bacteriocins display different modes of action such as protoplasm vesicularization, pore formation or cell disintegration (Sumi et al., 2015). They are FIGURE 1 | Timeline emergence of the species from the B. subtilis group. The species are classified following their relatedness to the closest original member of the group (gray boxes). Heterotypic synonyms are not shown.
FIGURE 2 | Antimicrobial molecules classes from the B. subtilis group. The subdivision between the classes is based on the biosynthetic pathway (i.e., ribosomal peptides, polyketides, hybrids, non-ribosomal peptides, and volatile compounds).
generally bactericidal with some exceptions that exhibit bacteriostatic activities (Gautam and Sharma, 2009). For most class I and II bacteriocins, the target of their activity is the bacterial envelope due to their amphiphilic or hydrophobic properties. For instance, lantibiotics from subclass I.1 have a dual mode of action. On the one hand, they can inhibit the cell wall synthesis of the targeted bacteria through binding to lipid II, the major transporter of peptidoglycan subunits FIGURE 3 | Lanthionine biosynthesis. General pathway of the lanthionine synthesis (A), structure of subtilin (B) and nisin A (C). Non-modified AA are indicated in teal whereas dehydrated serine (Dha, dehydroalanine) and threonine (Dhb, dehydrobutyrine) are colored in orange. The lanthionine (Ala-S-Ala, alanine-S-alanine) and R-methyllanthionine (Abu-S-Ala, aminobutyrate-S-alanine) bridges are shown in purple. The AA of nisin that differ from those in subtilin are highlighted as hatched circles. Adapted from Cotter et al. (2005) and Spieß et al. (2015).
across the inner cell membrane. On the other hand, lipid II can be used as a docking molecule to insert the lantibiotic in the membrane leading to pore formation and ultimately to cell death as well described in Chatterjee et al. (2005) and Cotter et al. (2005). This duality has been reported for subtilin, a class I bacteriocin which is active against a broad range of Grampositive bacteria such as Staphylococcus simulans, B. subtilis, and Bacillus stearothermophilus (Linnett and Strominger, 1973;Parisot et al., 2008).
Many regulation systems mediate bacteriocin production, secretion and immunity. Bacteriocin production is usually linked to particular cellular events such as stress responses. For instance, subtilin production depends on cell density and is increased under starvation conditions (Abriouel et al., 2011). Lantibiotic production is also mediated by QS. For subtilin, it has been demonstrated that the peptide itself acts as an auto-inducer of its own production (Kleerebezem, 2004). The export of bacteriocins is generally ensured by a dedicated membraneassociated ATP-Binding Cassette (ABC) transporter. For some lantibiotics, the cleavage of the leader peptide often occurs in a proteolytic domain present in the ABC transporter as described in McAuliffe et al. (2001) and Cotter et al. (2005). The immunity of the producing strains to its own active bacteriocin(s) can be achieved by several mechanisms like the secretion of immunity proteins sequestering the peptide, the bacteriocin re-export through an ABC transporter system or the alteration of the targeted peptidoglycans bonds (e.g., modification of the cell wall or cytoplasmic membrane charge) (Cotter et al., 2005;Dubois et al., 2009).

B. subtilis Group AMP Enzymes
Among the B. subtilis group, two major types of enzymes exhibit antagonistic activities (Supplementary Table S1): the lytic enzymes and those involved in quorum quenching (QQ). Several strains from the B. subtilis group have indeed been identified as capable to produce lytic enzymes with biocontrol potential (Herrera-Estrella and Chet, 1999;Kumar et al., 2012;Shafi et al., 2017). They include cellulases, glucanases, proteases and chitinases and are generally referred to as cell wall degrading enzymes (CWDE) (Ariffin et al., 2006;Alamri, 2015;Caulier et al., 2018). They are particularly active against fungi since chitin and glucan are the major constituents of their cell wall where various glycoproteins are embedded (Bowman and Free, 2006;Geraldine et al., 2013;Gomaa, 2012). Quorum quenching is able to silence or block QS which is generally defined as the cell-to-cell communication mechanism through the production of signal molecules (Czajkowski and Jafra, 2009). N-acyl-homoserine lactones (AHLs), composed of a fatty acid side chain and a homoserine lactone (Figure 4) are the most characterized signal autoinducers in Gram-negative bacteria. When a bacterial population proliferates, concentration of AHLs increases so that all the cells coordinate their metabolic activities (e.g., biofilm formation, sporulation, virulence factors or antibiotic production) (Dong et al., 2004). As the QS system brings ecological advantages to a coordinate population, QQ is able to counteract QS. Four types of enzymes (i.e., lactonase, decarboxylase, acylase, and deaminase) are able to inactivate AHLs, as illustrated in Figure 4 (Czajkowski and Jafra, 2009). B. subtilis AHL-lactonases have for instance attracted interest for biocontrol since they affect the growth of deleterious microbial pest such as Pectobacterium carotovorum subsp. carotovorum causing potato soft rot (González and Keshavan, 2006).

POLYKETIDES
Among the bioactive compounds produced by microorganisms, PKs are well known from the human health sector for their broad spectrum of activity encompassing antibacterial, immunosuppressive, antitumor and many more antagonistic abilities. Typical PKSs structures from the B. subtilis group are presented in Figure 5. They are synthetized from acyl CoA precursors such as malonate and methyl malonate. Their biosynthesis depends on multifunctional polyketide synthases (PKSs). Their structure was first extrapolated from fatty acid synthases (FASs) that share similarities in terms of chain extension mechanisms, precursors and overall architecture design (Smith and Tsai, 2007). As shown in Figure 6A, PKS are composed of a succession of elongation modules, flanked by initiation and termination modules. The reactive mechanism of these three PKS domains is illustrated in Figure 7A and is well summarized in Hertweck (2009). The initiation module is composed of two domains: an acyltransferase (AT) domain that recruits and catalyzes the binding of a monomer substrate to an acyl carrier protein (ACP) domain. The ACP then acts as an arm with a second catalytic domain located on the next elongation module. This domain, a β-ketoacyl synthase (KS), catalyzes the chain-elongation reaction that occurs through a decarboxylative Claisen thioester condensation (Cane and Walsh, 1999;Hertweck, 2009). In addition to the three core domains, auxiliary domains can also be present on elongation modules (gray domains in Figure 6A). These auxiliary domains mediate ketoreduction (KR), dehydration (DH), or enoylacyl reduction (ER) occurring before the chain-elongation reaction. These modifications considerably enrich the structural complexity and diversity of mature PKs (Hertweck, 2009). Finally, a termination module harboring an additional thiosterase (TE) domain catalyzes the macrolactonization and the release of the mature PK (Cane and Walsh, 1999).
Polyketide synthases have been classified in three canonical types based on the structural organization of their functional domains. Type I PKSs involve large multifunctional enzymes housing several domains linearly arranged and covalently bonded. Type II PKSs are multienzyme complexes composed of separate monofunctional enzymes combined during the PK synthesis. Type III PKSs are chalcone synthase-like PKSs that operate the acid CoA thioesters directly without any ACP domain (Chen and Du, 2016). Beside these structural differences, PKSs are classified as iterative or non-iterative depending on how many KS domains are used in the biosynthetic process. Within prokaryotes, the non-iterative type I PKSs is the most represented. They produce PK compounds that harbor a oneto-one correspondence with the PKS modular architecture. This conservation of collinearity is used for PKS discovery via genome mining (Challis, 2008).
Due to the diversity of PKSs, many exceptions and transition states between the three main types are observed. In some cases, mixed PKs pathways combine different types of PKSs or can even be associated with FASs or NRP synthetases (NRPSs) to form PKpeptide hybrid metabolites such as bacillaene, compactin, fusarin C or salinosporamide A (Moldenhauer et al., 2007;Hertweck, 2009;Fisch, 2013).   Cane and Walsh (1999) and Challis and Naismith (2004).

Polyketide class
Structure description Typical core unit or example structure Reference Acetogenins Linear 32-or 34-carbon chains with oxygenated functional groups bearing a terminal γ-lactone ring Li et al., 2008 Ansamycins Bridge between an aromatic moiety and an aliphatic chain Williams, 1975 Enediynes Compounds characterized by a core structure formed by a double C-C bond conjugated to two acetylenic groups Horsman et al., 2009 Macrolides Large macrocyclic lactone ring with one or more deoxy sugars

Yuan et al., 2012a
Polyenes Poly-unsaturated organic compounds containing at least three alterning single and double C-C bonds Hertweck, 2009 Polyethers Polymers containing more than one ether group Hertweck, 2009 Tetracyclines Compounds family characterized by a typical four-ring system Rohr, 1992 To date, seven PKs families have been recognized based on their carbon skeletons and typical structures, as summarized in Table 2 (Eustáquio et al., 2009). However, to our knowledge, only three antimicrobial PKs and their variants are produced within the B. subtilis group: bacillaene, difficidin, and macrolactin. These compounds exhibit antibacterial activities through selective inhibition of protein synthesis ( Table 3). Bacillaene is a polyene PK resulting from a hybrid synthesis by a type I PKS and a NRPS bae operon (baeJ, baeL, baeM, baeN and baeR) (Chen et al., 2006;Moldenhauer et al., 2007). Its exhibits antimicrobial activity against various bacteria (e.g., Myxococcus xanthus or Staphylococcus aureus) and fungi (e.g., Trichoderma spp. or Fusarium spp.) (Patel et al., 1995;Um et al., 2013;Müller et al., 2014). Difficidin, and its oxidized form oxydifficidin, c Activity of isolated compound confirmed by compound purification or mutant deletion, p putative activity of the compound contained in a broth mixture. * Two PKs classes are reported in this review (macrolides and polyenes) as well as the hybrids between PKs and NRPs. * * −, no activity known.

NON-RIBOSOMAL PEPTIDES
Non-ribosomal peptides form a versatile family of secondary metabolites with growing interest in many industrial fields as antibiotics, siderophores, surfactants, pigments, immunosuppressors or antitumor molecules . NRPs show a broad structural diversity, from linear to cyclic or branched structures (Kopp and Marahiel, 2007). As illustration, the Norine database counts almost 1.200 NRP molecules, including their structure, synthesis and evolution 2 (last update in January 2019) (Caboche et al., 2008). Two categories of NRPs can be distinguished whether they are synthetized through a multi-enzyme thio-template mechanism or not (Sumi et al., 2015). The first ones usually result in structures with two to ca. 50 residues and other moieties such as fatty acid chains [i.e., lipopeptides (LPs) and siderophores] whereas the second ones are generally smaller. Figure 8 shows the chemical structures of typical NRPS from the B. subtilis group.

Thiotemplate NRPs -Lipopeptides
Lipopeptides are usually synthetized through a NRPS sequential addition of AA residues, either in an iterative or noniterative way. Similarly to PKSs, NRPSs have a modular organization implementing the initiation, elongation, and termination modules ( Figure 6B). Each module is subdivided in core domains whose catalytic and carrier domains slightly differ from PKSs, as shown in Figure 7B. The biosynthesis which was previously summarized in Ongena and Jacques (2008) and Raaijmakers et al. (2010) starts with an adenylation domain (A domain) that recruits and phosphorylates an AA monomer into an aminoacyl adenylate intermediate. The intermediate is then linked to the corresponding peptidyl carrier protein or thiolation domain (PCP or T domain) through a thioester bond. The PCP acts as a bridge and ensures the link with the condensation domain (C domain) that forms the C-N bond between the recruited aminoacyl and the peptide acyl chain in formation. The termination module contains a thioesterase domain (TE) that catalyzes the release of the final peptide acyl chain (Ongena and Jacques, 2008;Raaijmakers et al., 2010). The elongation modules can be supplemented with accessory domains such as cyclization domain (Cy), epimerization domain (E) and methylation domain (M). Those domains are able to modify the growing peptide chain which leads to diverse mature compounds structure (Cane and Walsh, 1999;Challis and Naismith, 2004).
Since the LP biosynthetic pathways are highly flexible, the range of produced LPs is extremely heterogeneous. Among LPs produced by Bacillus spp., four main families have been distinguished: kurstakins, surfactins, iturins, and fengycins (Jacques, 2011). Each family shares the same structural features based on the nature and organization of the peptide moiety or fatty acid tail, as summarized in Table 4. Strains from the B. subtilis group produce surfactins, iturins and fengycins whereas kurstakins are produced by B. thuringiensis strains (Béchet et al., 2012). Among the three LP families produced by B. subtilis, at least eight fengycins, 13 surfactins and 14 iturins variants have been described so far, as detailed in Supplementary Table S2.
The mere composition of LPs, where a peptide moiety is bound to a lipid tail, gives them an amphiphilic property. This nature makes them excellent surfactants and plays a significant role in their biological functions and antimicrobial properties. Indeed, LPs are able to destabilize the plasma membrane via a pore forming activity leading to the cell death of the target microbes. Their antiviral activity is the result of a similar disintegration of the bi-lipid envelope of virions explaining the weak LPs activity against plant viruses among which very few are enveloped (Ongena and Jacques, 2008).
Bacillus spp. LPs have many other biological and ecological functions as fully documented by Raaijmakers et al. (2010). They are also known to impact other metabolic mechanisms such as biofilm formation, motility, virulence, plant root colonization, and plant defenses. Moreover, it has been suggested that their participation to the degradation of hydrophobic substrates could be used for polluted soils bioremediation (Mulligan et al., 2001). Although some lipopetides have already been exploited as food biopreservatives or crop protection products, the industrial interest for LPs in specific applications is unsurprisingly continuously growing.

Thiotemplate NRPs -Siderophore
Itoic acid is a mono-peptide composed of a 2,3-dihydroxybenzoate (DHB) molecule bound to a glycine. It is used as a precursor by trimodular NRPS machinery to produce bacillibactin which is obtained after a condensation of three units of DHB-glycine-threonine (May et al., 2001). The synthesis of the final hexapeptide is catalyzed by a terminal thioesterase domain leading to the production of a methylated trilactone ring link to three catecholates moieties. It is this cyclic structure that enables the sequestration of the metal atom (Dertz et al., 2006). Itoic acid and bacillibactin are both catecholic siderophores that chelates iron reducing its bioavailability. This is limited access to iron that allows B. subtilis to antagonize the growth of other surrounding microbes such as, for instance, F. oxysporum f. sp. capsici (Yu et al., 2011).

Non-thiotemplate NRPs
Bacteria from the B. subtilis group are also able to synthesize other antimicrobial NRPs through non-thiotemplate mechanism. Rhizocticins are di-and tri-phosphono-peptides. They are constituted of a L-2-amino-5-phosphono-3-cis-pentenoic acid (APPA) linked to an arginine (rhizocticin A). They can be supplemented with an additional valine (rhizocticin B), isoleucine (rhizocticine C) or leucine (rhizocticine D). After their integration into the target microbes, their cleavage by host cell peptidases releases the fungitoxic L-APPA moiety that interferes with threonine metabolism in fungal cells. Interestingly, rhizocticin A has also an antagonistic activity against nematodes such as Caenorhabditis elegans (Kugler et al., 1990).
In addition to rhizocticin compounds, two other dipeptide NRPs are produced by B. subtilis: bacilysin (also known as tetaine) and its chlorinated derivative, chlorotetain. They contain L-alanine (or chlorine-L-alanine) bound to the nonproteinogenic L-anticapsin Rapp et al., 1988). Despite their simple composition, these bioactive compounds display strong antibacterial activity mediated by the anticapsin moiety that inhibits the glucosamine-6phosphate synthase. Its inhibition suppresses the biosynthesis of peptidoglycans that are the main constituents of bacterial cell wall (Steinborn et al., 2005;Mahlstedt and Walsh, 2010). For the fungi, it has been proposed that because anticapsin is able to inhibit the production of chitin and fungal membrane mannoproteins, bacilysin and chlorotetain exhibit antifungal activity against Aspergillus fumigatus or C. albicans (Milewski et al., 1986;Rapp et al., 1988).
Finally, bacitracin and mycobacillin are two cyclic polypeptides produced by B. subtilis. Bacitracins are dodecapeptides containing a cyclic heptapeptide linked to a thiazoline ring (Johnson et al., 1945). They are mostly active against Gram-positive bacteria where they inhibit the bacterial cell-wall biosynthesis by preventing the lipid carrier from re-entering in the reaction cycle of peptidoglycan synthesis (Siewert and Strominger, 1967). Besides this primary mode of action, bacitracin might also act through other mechanisms affecting membrane functions, hydrolytic enzymes and/or the biosynthesis of ubiquinone precursors (Konz et al., 1997). Mycobacillin is an antifungal cyclic tridecapeptide altering the membrane of fungi like Aspergillus niger (Majumdar and Bose, 1958). Interestingly, its biosynthesis is rather peculiar. Although it is catalyzed by a large NRPS complex, it is divided in three fractions (A, B, and C) and does not use a thio-template mechanism (Zuber et al., 1993). Each fraction of the enzymatic complex contains a single enzyme polypeptide that catalyzes the polymerization of a first pentapeptide (A), a second nonapeptide (B) and the final tridecapeptide.

VOLATILES
Besides RPs, NRPs and PKs, strains from the B. subtilis group are able to produce a wide diversity of volatile compounds encompassing important roles especially in soil, one of the major habitats of this group (Supplementary Figure S1). Volatiles are notably involved in the bioconversion of the food chain, in the biogeochemical cycles of essential elements, in many physiological and metabolic reactions (e.g., nitrification, nitrogen mineralization, electron acceptor or donor reactions) as well as in communication signals triggering QS/QQ or defense mechanisms well reviewed in Effmert et al. (2012). Volatile compounds are generally classified into inorganic (VICs) and organic (VOCs) categories.

Volatile Inorganic Compounds (VICs)
Volatile inorganic compounds synthesized by microorganisms are mainly by-products of primary metabolism. They are carbonated, hydrogenated, sulfur or nitrogen-containing compounds such as CO 2 , CO, H 2 , HCN, H 2 S, N 2 , NH 3 and NO. Nitrogen-containing compounds are mostly released in aerated upper sediments layers by denitrifying bacteria. In this process, nitric oxide is enzymatically produced by the nitric-oxide reductase or the nitric-oxide synthase (Adak et al., 2002). The range of antimicrobial activities exhibited by VIC nitrogen-containing compounds from the B. subtilis group is wide. For instance, NO is able to induce systemic acquired resistance (SAR) in plants against bacterial pathogens such as R. solanacearum (Wang et al., 2005). A contrario, ammonia, a secondary metabolite from the catabolism of the amino acids L-aspartate, is known to be active against soil-borne Oomycetes such as Pythium spp. (Howell et al., 1988). Hydrogen cyanide, derived from the glycine catabolism, shows a direct antagonistic activity against aerobic microorganisms by inhibiting metalcontaining enzymes such as the cytochrome c oxidase active in the respiration chain (Cherif-Silini et al., 2016).
Deeper in the soil, under low oxygen concentration, bacteria tend to produce different VICs such as H 2 or H 2 S. Those compounds can serve as electron acceptors, AA precursors or antimicrobial metabolites. Hydrogen sulfide could be produced by B. subtilis from sulfate reduction or as a by-product of L-methionine and L-cysteine catabolism via a direct cleavage of L-methionine or a transamination followed by reductive demethiolations (Even et al., 2006;Schulz and Dickschat, 2007). It is known to exhibit antifungal activity against several plant pathogens such as A. niger or Penicillium italicum but also against some food-borne bacteria or human pathogens (Fu et al., 2014). Curiously, it is also known to act as a bacterial defense mechanism against antibiotics (Shatalin et al., 2011). Interestingly, ammonia increases the resistance of several Gram-negative and Grampositive bacteria to antibiotics too (Bernier et al., 2011).

Volatile Organic Compounds (VOCs)
Volatile organic compounds are small compounds with fewer than 20 carbon atoms and are characterized by low molecular mass (100-500 Da), high vapor pressure, low boiling point and a lipophilic moiety. These features ensure an easy evaporation and a long distance distribution which is convenient in a complex matrix like soil (Schmidt et al., 2015). Their diffusion and production by soil-borne microbes are strongly dependent on various factors such as nutrient and oxygen availability, temperature, pH, physiological state of microorganisms, soil moisture, texture and architecture (McNeal and Herbert, 2009;Insam and Seewald, 2010;Effmert et al., 2012). The majority of VOCs derives from glucose oxidation involving glycolysis and the subsequent cycles such as the tricarboxylic acid cycle (TCA) as it has been well summarized in Korpi et al. (2009) andSchmidt et al. (2015). However, their production can also result from various other pathways such as aerobic heterotrophic carbon metabolism, fermentations, AA degradation, terpenes synthesis or sulfur reduction (Peñuelas et al., 2014). Based on previous reviews presented in Schulz and Dickschat (2007); Peñuelas et al. (2014) and Audrain et al. (2015), five categories of VOCs can be distinguished: (1) fatty acids and derivatives, (2) terpenoids, (3) nitrogen-containing VOCs, (4) sulfur-containing VOCs, and (5) metalloid-or halogenated-containing VOCs. To date, about 2,000 compounds produced by almost 1,000 species of microorganisms have been listed in the mVOC 2.0 database (Lemfack et al., 2018). According to this database, almost 70% of recorded Bacillus VOCs are fatty acids derivatives (alcohols, ketones, alkanes, aldehydes, alkenes, and acids) followed by sulfur-and nitrogen-containing compounds. Supplementary  Table S3 displays the VOCs produced within the B. subtilis group and their antimicrobial activity.
Since many volatile fatty acids and their derivatives result from the glucose metabolism, their precursors mostly derive from the Embden-Meyerhof (glycolysis), Entner-Doudoroff, heterolactic and homolactic fermentation pathways (Peñuelas et al., 2014). B. subtilis bacteria, for instance, ferment pyruvate to produce ketone compounds such as acetoin (3-hydroxy-2-butanone) or 2,3-butanedione under anaerobic conditions (Ryu et al., 2003). Other intermediates coming from fatty acid biosyntheses or their β-oxydations are also used as precursors by microbes and transformed into VOCs through a decarboxylation reaction or a reduction of their carboxyl group (Schulz and Dickschat, 2007). They provide essential hydrocarbons but also other fatty acid derivatives. An oxidative deamination of several amino acids can lead to the production of aldehyde, ketone or alcohol volatile too. For instance, the degradation of L-phenylalanine or L-tyrosine can be the first step of the aromatic volatile compounds synthesis such as benzene or its carbohydrate derivatives. Finally, benzenoid volatiles can also be synthesized by microbes through the shikimate pathway that leads to the formation of chorismate, a natural precursor of aromatic amino acids (Bentley and Haslam, 1990). Degradation of intermediates from the shikimate pathway or aromatic amino acids can also lead to the production of benzenoid volatiles (Dickschat et al., 2005).
This wide variety of volatile fatty acids and their derivatives make them the most important group of VOCs produce by microbes and represent up to 87% of known antimicrobial VOCs produced by B. subtilis bacteria (Supplementary Table S3). They can be divided in two main categories: hydrocarbons (alkanes, alkenes, alkynes) or carbohydrates (acids, alcohols, aldehydes, esters, furans, ketones, lactones, benzenoids). Among them, benzenoids is the most represented sub-category followed by alkanes, aldehydes, ketones, acids, and alcohols. Even though benzenoids could be considered as an individual category, they can also be seen as fatty acids derivatives because a large majority of antimicrobial benzenoid volatile produced by B. subtilis harbor a benzene core linked to a fatty acid derivatives.
There is an important diversity of benzenoids, sometimes linked with carbohydrate chains containing nitrogen, sulfur or both. Most of these antimicrobial volatile exert fungicidal activities but some have been characterized for their antibacterial or nematicidal abilities, too. Their mode of action is rarely fully characterized. For instance, morphological abnormalities on fungal and bacterial cells have been documented after an exposition to B. subtilis VOCs (Tahir et al., 2017). Volatile such as 1,3-butadiene or 2,3-butanediol are also known to induce modifications in the expression of genes linked to the pathogenicity of R. solanacearum and Pectobacterium carotovorum (Marquez-Villavicencio et al., 2011;Tahir et al., 2017). In addition to direct antimicrobial activities, fatty acids volatile have also several other biological functions. For instance, acetoin and 2-butanone have the ability to stimulate plant defenses or to induce plant stress tolerance which then promote plant growth (Ryu et al., 2003;Ryu et al., 2004;Ryu, 2015). They are essentially produced by strains of B. amyloliquefaciens, B. velezensis or B. subtilis (Audrain et al., 2015).
Terpenes and their derivatives (also known as terpenoids or isoprenoids) are among the most abundant secondary metabolites found in living systems (Fisher et al., 2001;Gershenzon and Dudareva, 2007). They originate from two main precursors: isopentenyl pyrophosphate (IPP) and its allylic isomer the dimethylallyl pyrophosphate (DMAPP) (Schulz and Dickschat, 2007). IPP and DMAPP are also the end-products of the deoxy-xylulose phosphate pathway (DOXP) starting with pyruvate and glyceraldehyde-3-phosphate originating from the glucose metabolism (Fisher et al., 2001). Terpenoids can be synthesized from isoprene molecules too. Julsing et al. (2007) showed that, in B. subtilis, isoprene is not formed by the MVA or DOXP pathways but, as in plant systems, might be a product of the methylerythritol phosphate (MEP) pathway (Guan et al., 2015).
Isoprenoid compounds are produced by all living organisms for essential physiological functions such as electron transport, membrane fluidity, light harvesting, photoprotection, anchoring of molecules to specific membranes and signaling (Fisher et al., 2001). The signaling ability is particularly important and is associated with several antagonistic, mutualistic or multi-trophic interactions (Shrivastava et al., 2015). More than 25,000 terpenic compounds have been listed and, for the vast majority, their biological functions and roles remain unknown (Buckingham, 1997). Volatile terpenes are generally recognized for their ability to inhibit bacteria (Scortichini and Rossi, 1991), fungi (Hammer et al., 2003;Dambolena et al., 2008), nematodes (Gu et al., 2007) or insects (Lee et al., 2003;Justicia et al., 2005). They can be classified in three categories: isoprene, monoterpenes (C 10 ) and sesquiterpenes (C 15 ) (Schmidt et al., 2015).
The mode of action of these compounds might be linked to their lipophilic nature allowing them to destabilize the cell membrane integrity (Cox et al., 2000;Inoue et al., 2004). To our knowledge, only two terpenes produced by B. subtilis show antimicrobial abilities: isoprene and monoterpene α-terpineol exhibit antagonistic activities against cyanobacteria and nematodes (Wright and Thompson, 1985;Gu et al., 2007).
Little is known about the biosynthetic pathways of nitrogencontaining VOCs. Nevertheless, it is accepted that two main routes can be used: a non-enzymatic amination of acyloins, that can lead to the formation of pyrazines (Schulz and Dickschat, 2007) or derived from α-aminoketone intermediates resulting from AA catabolism (Owens et al., 1997;Zhu et al., 2010).
Nitrogen-containing VOCs can be distinguished based on their cyclization rate. Within non-cyclic compounds, three groups are identified (amides, amines and imines) while there are five categories of cyclic compounds (azoles, pyrazines, pyridines, pyridazines, and pyrimidines). Pyrazines are strongly represented among microbial volatile and are separated in two classes: loweralkylated and higher-alkylated pyrazines (Schulz and Dickschat, 2007). These compounds are characterized by a strong odor and several B. subtilis coming from the rhizosphere or from food fermentations have already been recognized as pyrazines producers (Sugawara et al., 1985;Kosuge and Kamiya, 1962;Larroche et al., 1999;Leejeerajumnean et al., 2001). Pyrazines from B. subtilis strains are known to exhibit antifungal and nematicidal activities (Gu et al., 2007;Chen et al., 2008;Chaves-López et al., 2015;Haidar et al., 2016). For instance, tetramethylpyrazine inhibits the growth of Moniliophthora perniciosa and F. oxysporum f. sp. lactucae. Additionally, it acts on sporulation and elongation of the germ-tube of B. cinerea Chaves-López et al., 2015). It is interesting to note that B. subtilis pyrazines can also exhibit antibacterial activities such as pulcherriminic acid which inhibits the growth of S. aureus, E. coli and Proteus vulgaris (Coutts et al., 1965). Beside pyrazines, strains from the B. subtilis group are able to produce other nitrogen VOCs such as 1H-imidazole,1-ethyl showing antifungal activities against numerous soil-borne phytopathogens (Lupetti et al., 2002;Liu et al., 2008;Snelders et al., 2009;Schmidt et al., 2015).
Other volatile organic compounds such as halogenated, metalloids, tellurium or selenium compounds have also been described. However, at the time of writing, no B. subtilis strains have been proved to produce these type of VOCs (Schulz and Dickschat, 2007), although related bacteria, like Bacillus arsenicoselenatis, have been shown to generate them (Switzer Blum et al., 1998).

CONCLUSION AND PERSPECTIVES
The B. subtilis group offers a plethora of antagonistic compounds displaying a broad range of biological functions. This huge versatility increases the industrial and environmental interest of B. subtilis strains, especially when considering their range of action against foodborne or phytopathogenic flora as well as their history of safe use in food. The present review on known AMCs from the B. subtilis group proposes a consistent classification frame based on their biosynthetic pathways (i.e., RPs, PKs, NRPs, volatiles) and chemical nature.
The present classification suggests to establish systematic approaches for novel molecules discoveries and characterizations (biosynthesis, chemical nature and activity). Indeed, most current publications report antimicrobial activity of partially purified fractions which can involve mixtures of bioactive compounds. To assess the activity of an unique compound, implementations of genetic confirmation such as knockout strategy are needed. Besides, very few studies have focused on the putative synergistic effects within these bio-active mixtures. Also, the concentration of purified or semi-purified compound(s) often remains uncharacterized or biologically irrelevant. Finally, there is no doubt that novel AMCs originating from B. subtilis bacteria remain to be identified, characterized and properly classified.

AUTHOR CONTRIBUTIONS
SC, CN, and FL conducted the bibliographic search. SC and CN wrote the manuscript. AG, CB, and JM edited and reviewed the manuscript. All authors have read and approved the final version.

FUNDING
This work was supported by the National Fund for Scientific Research (FNRS), the Université catholique de Louvain (UCLouvain), and the Brussels Institute for Research and Innovation (Innoviris, Doctiris programme to CN). SC was supported by the Foundation for Training in Industrial and Agricultural Research (FRIA, FNRS), AG holds a Chargé de Recherche fellowship from the FNRS (Grant 1.B.208.16F).

ACKNOWLEDGMENTS
We gratefully acknowledge members of SC's Ph.D. committee Prof A. Legrève, UCLouvain, Prof. M. Ongena, ULiège and Dr. J.-P. Goffart, CRAw for their valuable comments on this manuscript. We also acknowledge the Walloon Region for the long-term financial support through the WACOBI and ANTAGONIST projects (Conventions N • DGO3-D31-1330 and N • DGO3-D31-1383/S1, respectively).