New Insight into Antimicrobial Compounds from Food and Marine-sourced Carnobacterium Species through Phenotype and Genome Analyses

Carnobacteriummaltaromaticum and Carnobacterium divergens, isolated from food products, are lactic acid bacteria known to produce active and efficient bacteriocins. Other species, particularly those originating from marine sources, are less studied. The aim of the study is to select promising strains with antimicrobial potential by combining genomic and phenotypic approaches on large datasets comprising 12 Carnobacterium species. The biosynthetic gene cluster (BGCs) diversity of 39 publicly available Carnobacterium spp. genomes revealed 67 BGCs, distributed according to the species and ecological niches. From zero to six BGCs were predicted per strain and classified into four classes: terpene, NRPS (non-ribosomal peptide synthetase), NRPS-PKS (hybrid non-ribosomal peptide synthetase-polyketide synthase), RiPP (ribosomally synthesized and post-translationally modified peptide). In parallel, the antimicrobial activity of 260 strains from seafood products was evaluated. Among the 60% of active strains, three genomes were sequenced and submitted to a dereplication process. C. inhibens MIP2551 produced a high amountof H2O2, probably thanks to the presence of four oxidase-encoding genes. C. maltaromaticum EBP3019 and SF668 strains were highly efficient against Listeria monocytogenes. A new extracellular 16 kDa unmodified bacteriocin in the EBP3019 strain and five different bacteriocins in SF668 were highlighted. In this study, the overview of antimicrobial BGC and inhibitory activities of Carnobacterium spp. allowed the prediction of potential innovative natural products that could be relevant for biotechnological applications.


Introduction
Lactic acid bacteria (LAB) are microaerophilic Gram-positive bacteria capable of fermenting sugars into lactic acid. Due to this feature, LAB are involved in various food fermentations such as dairy, meat, or vegetable products. Moreover, they are ubiquitous microorganisms present in many terrestrial environments, ranging from soil, plants, or animals. LAB can also be found in marine environments such as coastal and estuarine sediment or in fish gastrointestinal tract [1]. LAB have developed strategies to outcompete microorganisms in these various ecosystems, such as nutritional competition, environmental acidification, or production of secondary metabolites [2].

BGC Prediction
All Carnobacterium spp. genomes were submitted to the RAST nmpdr (Rapid Annotation using Subsystems technology) [36] and MicroScope platform for automatic annotation [35].
BGCs putatively involved in antimicrobial compound biosyntheses were identified by combining the AntiSMASH 5.0 [39] and BAGEL 4 [40] softwares. AntiSMASH is a web-based pipeline able to predict a large diversity of BGCs such as RiPPs, NRPS, PKS, terpenes, and siderophores. BAGEL is specialized in bacteriocin and RiPP cluster prediction. For each predicted BGC class, the presence of a minimal gene subset required for BGCs was investigated as follows. For RiPP clusters, the presence of a structural gene (core gene), an immunity protein-encoding gene, and accessory genes leading to post-translational modifications was investigated. When the core or immunity protein gene was not predicted, it was manually searched using Artemis [41] to predict small coding sequences (CDS). BLASTp was used to compare core sequences, conserved domains, and genetic environments of the Carnobacterium spp. BGCs. The hydrophobicity and high isoelectric point for the predicted head to tail cyclized peptide core sequence were checked according to Gabrielsen et al. with GPMAW lite [42,43]. SignalP-5.0 was used to predict extracellular proteins and the cleavage position between the peptide signal and the core peptide [44]. For NRPS/PKS clusters identified by AntiSMASH, the presence of the required initiation and termination modules was checked.
A BGC network was built using the BiG-SCAPE software [45]. This network includes 77 BGCs previously predicted with AntiSMASH and BAGEL. When available, the genetically closest BGCs, already described in the literature were included. The network was built with a local mode and a cutoff parameter of 0.7. The layout of the network was done with BiG-SCAPE for spacing and connections between the different nodes and with Cytoscape for colorization and connector width based on the squared similarity parameter.

Carnobacterium spp. Strains, Growth Conditions, and Identification
For this study, 260 Carnobacterium spp. strains isolated by Ifremer (Nantes, France) from seafood products (smoked salmon, shrimps, and cod), and 12 reference strains were used (Table S3). Strains were stored at −80 • C in 15% (v/v) glycerol. Two successive pre-cultures were performed in BHI medium at 26 • C for 24 h.
Strains were identified using partial 16S rDNA sequencing. Briefly, 500 µL of culture suspensions were centrifuged for 20 min at 4500 g. Pellets were washed twice in 500 µL of 10 mM Tris containing 1 mM EDTA and resuspended in 200 µL of the same buffer before heating for 10 min at 95 • C. After centrifugation for 10 min at 4500× g, 1 µL of supernatant was used for PCR amplification of the 16S rRNA gene using the universal primers, E8-F (5 -AGAGTTTGATCATGGCTCAG-3 ) and 1489R (5 -GTTACCTTGTTACGACTTCAC-3 ). PCR amplification was performed under the following conditions: initial denaturation at 95 • C for 5 min; 30 cycles of amplification, including three steps (95 • C for 30 s, 52 • C for 30 s, 72 • C for 1 min), final extension for 10 min at 72 • C using a T100 TM Thermal Cycler (Bio-Rad, Hercules, CA, USA) in a final volume of 50 µL. The master mix 1X with the DreamTaq Green polymerase (Thermo Fisher Scientific, Waltham, MA, USA) was used. PCR products were purified, and the V1-V4 region was sequenced using the E8-F primer (Genoscreen, Lille, France).

Spot-On Lawn Assays
Carnobacterium spp. strains (260 from Ifremer collection and 12 references) were screened for antimicrobial activity by the spot-on lawn method adapted from Matamoros et al. [46]. Seventeen indicator strains involved in human infections, fish diseases, and food spoilage (Table S3)  Ten microliters of 24-h culture broth or cell-free supernatants (CFSs) were spotted on the surface of the soft agar medium seeded with one of the bacterial or yeast indicator strains. Indicator bacterial strains were prepared as follows-two successive cultures were performed in 10 mL appropriate medium at 26 • C for 24 h. Except for Aeromonas salmonicida, Chromobacterium violaceum, and Pseudomonas fluorescens, which were grown in Tryptic Soy Broth (TSB, Biokar Diagnostic, Beauvais, France), all bacteria grew in BHI medium. Candida albicans was cultured in Lysogeny Broth (LB, 1% Tryptone; 0.5% Yeast extract; 1% NaCl). Bacterial indicator strains and Candida albicans were then inoculated in 20 mL of the culture medium supplemented with 0.5% agar to reach an initial concentration of 10 4 CFU/mL and poured into one square Petri dish (120 × 120 mm). A 10 6 spores/mL Aspergillus fumigatus MMS839 suspension was prepared by scraping cells with sterile water from a 96-h culture on Dextrose Casein Agar medium (DCA, Detroit, MI, USA). Three milliliters of spore suspension was left for 5 min on a 0.5% agar Mueller Hinton Broth (DCA, Detroit, MI, USA) surface in a square Petri dish (120 × 120 mm). The excess of liquid was discarded, and then the Carnobacterium spp. cultures were spotted. Clear haloes observed after 24 or 48-h incubation at 26 • C indicated growth inhibition.

Hydrogen Peroxide Quantification
In order to assess the inhibition caused by the production of hydrogen peroxide (H 2 O 2 ), antimicrobial assays were performed by adding 5000 Units of bovine liver catalase (Merck, Darmstadt, Germany) in the agar medium. Carnobacterium spp. cultures were performed in 1 mL of BHI medium in 96 Deepwell plates for 24 h at 26 • C. After centrifugation for 30 min at 5000× g, 250 µL of supernatants were transferred into the Multiscreen TM system consisting of a microplate containing 0.22 µm polyvinylidene fluoride (PVDF) filters, Merck Millipore, Burlington, NJ, USA), a 96-well microplate receiver (Greiner Bio-One, Grosseron, France), and a Centrifuge Alignment Device (Merck Millipore, Burlington, NJ, USA). After 5 min of centrifugation at 5000 × g, cell free supernatants (CFSs) were collected and stored at −80 • C.
To determine whether the active compounds were peptidic, 100 µL CFSs were digested by 1 µL of proteinase K (Proteinase K from Engyodontium album EC 3.4.21.64; Merck, Darmstadt, Germany) at a final concentration of 0.2 mg/mL for 1 h at 37 • C. The digested CFSs were then spotted on previously inhibited indicator strains seeded in the agar medium, as described above. The absence of inhibition halo after proteinase K treatment evidenced an active antimicrobial peptide. The efficacy of each CFS was compared using L. monocytogenes RF191 as the indicator strain. Two-fold serial dilutions of CFS suspensions in BHI were inoculated with 10 6 CFU/mL of L. monocytogenes RF191 and incubated for 24 h at 26 • C. The titration of Carnobacterium spp. CFSs was carried out by measuring the growth of L. monocytogenes by OD 600 nm with a spectrophotometer (Varioskan TM , Thermo Fisher Scientific, Waltham, MA, USA). A negative control, CFS of Carnobacterium maltaromaticum EBP3034 strain, was included in the analysis. The minimal inhibitory dilution (MID) was determined by the highest dilution of CFS for which no growth of L. monocytogenes was measured.

Nucleotide Sequence Accession Number
The Whole Genome Shotgun projects were deposited in DDBJ/ENA/GenBank under the related accession number WNJQ00000000 for the MIP2551 strain, WNJR00000000 for the SF668 strain, and WNJS00000000 for the EBP3019 strain. The version described in this paper is the first one.

Carnobacterium spp. Genome Dataset
Eighty-nine Carnobacterium spp. genome sequences were retrieved from the NCBI database (Table S2). The species affiliation was checked by 16S rDNA-based phylogeny ( Figure S1) and by phylogenomics using the ANI (Average Nucleotide Identity) similarity values ( Figure 1). Whereas the analysis by 16S rDNA-based phylogeny alone was sufficient for most species' affiliation, in accordance with the current classification with a high threshold of 99%, the phylogenomics one was required to distinguish the species C. alterfunditum and C. pleistocenium. The Carnobacterium sp. AT7 strain was, thus identified as C. jeotgali, and the WN1374 and 17-4 strains as C. viridans species. This analysis also detected three erroneously identified Carnobacterium sp.-the strain ZWU0011 was identified in this study as Pisciglobus halotolerans, the strain 757_CMAL as Granulicatella adaciens, and the strain 1290_CSPC as Aerococcus urinaequii (Figure 1 and Figure S1). Consequently, these three strains were excluded from our dataset. To simplify further analysis, redundant genomes were also removed. Finally, 39 Carnobacterium spp. genomes were selected from the dataset to mine Carnobacterium spp. BGCs.

Diversity of Antimicrobial BGCs
A total of 67 manually annotated BGCs were considered as potentially functional since the required minimal gene set was displayed (Figures 2 and 3). At least one BGC could be predicted in 77% of the analyzed genomes ( Figure 2). Twenty-one different BGCs were identified; 14 of them were not described in the literature until now. They were distributed into four classes of natural products (Terpenes, RiPPs, NRPS, NRPS/PKS). Carnobacterium spp. strains harbored between zero and six BGCs. Two or three BGC copies were identified in scarce cases ( Figure 2). Globally, the BGC content seemed to be species-dependent. The water-sourced strains displayed specialized BGC content, but no significant difference was observed between the food strains isolated from seafood or meat products ( Figure 2).
Terpene RiPP Lanthipeptide Head to tail Unmodified bacteriocin   The network was generated with BiG-SCAPE, and graphical modifications were done with Cytoscape. A node represents a predicted BGC. For more legibility, several similar BGCs were grouped into a single node. In this case, a pie chart illustrates the represented species, and the size of the node is proportional to the number of BGCs. The thickness of the lines is correlated to the similarity between two nodes. The dotted boxes indicate the different classes of BGCs.

T 1 T 2 T 3 N R P S 1 A 1 / A 2 L a n 1 H T 1 H T 2 H T 3 T h 1 V 4 1 D iv e r g ic in
Terpenes are an important source of natural compounds derived from one or more isoprene unit. A total of 13 terpenes classified into three different groups (T1, T2, and T3) were predicted (Figures 2 and 3). The distribution of these terpenes appeared to be species-related and particularly distributed among environmental strains. Core clusters of T1 and T2 were only composed of, respectively, one and two biosynthetic genes related to terpenes ( Figure 4). In both T1 and T2, one gene coding for a putative phytoene/squalene synthase was predicted with a conserved isoprenyl diphosphate synthase domain (T1A, T2A). This enzyme is generally involved in the formation of the linear backbone of isoprenoid compounds [49]. T2B, an additional gene encoding a bacterial-type phytoene desaturase involved in tetraterpene biosynthesis such as carotenoids, was predicted in the cluster T2 [50]. The T3 BGC was more complex and was composed of six core biosynthetic genes, which code for proteins similar to those in the staphyloxanthin BGC from Staphylococcus aureus ( Figure 3) with the protein sequence identity varying between 38.3% and 58.9%. Staphyloxanthin is a carotenoid pigment promoting resistance to reactive oxygen species [51]. The network was generated with BiG-SCAPE, and graphical modifications were done with Cytoscape. A node represents a predicted BGC. For more legibility, several similar BGCs were grouped into a single node. In this case, a pie chart illustrates the represented species, and the size of the node is proportional to the number of BGCs. The thickness of the lines is correlated to the similarity between two nodes. The dotted boxes indicate the different classes of BGCs.
Terpenes are an important source of natural compounds derived from one or more isoprene unit. A total of 13 terpenes classified into three different groups (T1, T2, and T3) were predicted (Figures 2 and 3). The distribution of these terpenes appeared to be species-related and particularly distributed among environmental strains. Core clusters of T1 and T2 were only composed of, respectively, one and two biosynthetic genes related to terpenes ( Figure 4). In both T1 and T2, one gene coding for a putative phytoene/squalene synthase was predicted with a conserved isoprenyl diphosphate synthase domain (T1A, T2A). This enzyme is generally involved in the formation of the linear backbone of isoprenoid compounds [49]. T2B, an additional gene encoding a bacterial-type phytoene desaturase involved in tetraterpene biosynthesis such as carotenoids, was predicted in the cluster T2 [50]. The T3 BGC was more complex and was composed of six core biosynthetic genes, which code for proteins similar to those in the staphyloxanthin BGC from Staphylococcus aureus (Figure 3) with the protein sequence identity varying between 38.3% and 58.9%. Staphyloxanthin is a carotenoid pigment promoting resistance to reactive oxygen species [51].

288
Two different lanthipeptide BGCs were predicted (Figures 2 and 3). The two-component  In addition to the known NRPS/PKS hybrid BGC from C. divergens V41, only one NRPS BGC was detected (Figures 2 and 3). The new 43 kb-long NRPS cluster found in the C. inhibens DSM13024 strain (also named K1) was composed of an NRPS gene consisting of five complete domain modules (k1A) and of a second gene containing the termination module (k1B) [52] followed by a phosphopantethienyl transferase (k1C) [53] (Figure 4). Genes involved in transport or regulation (k1G-k1Y), as well as putative auxiliary genes, including a S-adenosyl-L-methionine (SAM)-dependent methyltransferase (k1D), an aspartate decarboxylase, and a phosphohydrolase (k1E and k1F) were also identified.
Two different lanthipeptide BGCs were predicted (Figures 2 and 3). The two-component lanthipeptide carnolysin A1/A2 [54] was detected in approximately half of the C. maltaromaticum strains (Figure 2). The amino acid sequence comparison revealed a 100% identity between all predicted canolysin A1/A2 BGCs. A new lanthipeptide BGC, named Lan1, was found in the C. divergens C13 strain. The cluster encompassed a lanM lanthionine synthetase gene, indicating a class II lanthipeptide [13], followed by three potential precursor peptides (Lan1A1, Lan1A2, and Lan1A3), and a transport system (Figure 4). The best similarity was found with the undescribed sequences from Bacillus thuringiensis with an amino acid identity below 68% (Table 1). Table 1. Undescribed ribosomally synthesized and post-translationally modified peptides (RiPPs) sequence prediction. Names were arbitrarily given depending on predicted RiPP classes. Sequence prediction was achieved with AntiSMASH, and the BAGEL leader sequence was predicted with SignalP 5.0. Alignment by Blastp of the core sequence to the non-redundant protein sequence database (nr) was used to predict the RiPP class and the closest known RiPP. Only one thiopeptide BGC was predicted in a half of the C. divergens strains (Th1) (Figure 2). Thiopeptides are RiPPs containing characteristic thiazole rings, described for the first time in Carnobacterium genus. The core cluster was composed of two structural genes, named th1A1 and th1A2, with identical sequences (Figure 4 and Table 1). Genes encoding enzymes involved in the structural backbone of the thiopeptide were predicted (e.g., formation of piperidine, dehydropiperidine and pyridine macrocycle, thiazole and thiazoline macrocycle, and Dha and Dhb residues) [13]. They included two LanB-like dehydratases (th1B and th1C), a potential enzyme involved in cycloaddition (th1D), and a cyclodehydratase (th1G). The genes involved in the regulation and transport were also predicted, including a putative cyclic autoinducer peptide (th1R). Th1 showed a 100% identity with the uncharacterized thiopeptide core sequence of Enterococcus termitis (Table 1).

Gene
Only one head to tail cyclized peptide (HT) named carnocyclin A has been reported in Carnobacterium species [55], but it was not found in any genome analyzed in this study (Figure 3). Three undescribed HT, namely HT1, HT2, and HT3, with original core sequences were predicted in C. maltaromaticum and C. viridans (Figure 2 and Table 1). Their molecular mass was estimated between 5.89 and 7.32 kDa.
No inhibition of Candida albicans and Pseudomonas fluorescens was observed. Approximately 60% of the Carnobacterium spp. strains showed antimicrobial activity against at least one of the indicator strains tested. Overall, different inhibition patterns were observed depending on the strain species ( Figure 5). C. maltaromaticum inhibited all Gram-positive species except Staphylococcus epidermidis. In contrast, other Carnobacterium species inhibited Escherichia coli, Vibrio parahaemolyticus, Morganella morganii, Staphylococcus epidermidis, and Aspergillus fumigatus. Interestingly, the C. inhibens MIP2551 strain was the sole strain able to inhibit Aeromonas salmonicida, Vibrio harveyi, and Chromobacterium violaceum.

357
The C. inhibens MIP2551 strain was the highest producer of H2O2; its production time-course was 358 monitored for 24 h and compared to Lactococcus garvieae, a species known to produce high inhibitory 359 amounts of H2O2 ( Figure 6). In optimal aeration conditions, the MIP2551 strain began to produce

Involvement of H 2 O 2 Inhibition
The addition of catalase into the medium resulted in the loss of the inhibitory effect for 56% of the active Carnobacterium spp. strains (Table S3). H 2 O 2 production is involved in the inhibition mechanism of S. epidermidis, E. coli, M. morganii and V. parahaemolyticus by C. viridans, C. mobile, C. pleistocenium, C. jeotgali, C. funditum, C. alterfunditum, C. inhibens, and by 67% of C. divergens strains. The C. inhibens MIP2551 strain was the highest producer of H 2 O 2 ; its production time-course was monitored for 24 h and compared to Lactococcus garvieae, a species known to produce high inhibitory amounts of H 2 O 2 ( Figure 6). In optimal aeration conditions, the MIP2551 strain began to produce H 2 O 2 during the exponential phase and reached a maximal concentration of 150 mg/L on the stationary phase, with a maximal OD 600nm of 0.3. This concentration was five times higher than those observed for Lactococcus garvieae strain. The same H 2 O 2 production profile was observed for C. inhibens CD344, DSM 13024, and WN1359 strains (data not shown).

Involvement of H2O2 Inhibition
The addition of catalase into the medium resulted in the loss of the inhibitory effect for 56% of the active Carnobacterium spp. strains (Table S3). H2O2 production is involved in the inhibition mechanism of S. epidermidis, E. coli, M. morganii and V. parahaemolyticus by C. viridans, C. mobile,  C. pleistocenium, C. jeotgali, C. funditum, C. alterfunditum, C. inhibens, and by 67% of C. divergens strains. The C. inhibens MIP2551 strain was the highest producer of H2O2; its production time-course was monitored for 24 h and compared to Lactococcus garvieae, a species known to produce high inhibitory amounts of H2O2 ( Figure 6). In optimal aeration conditions, the MIP2551 strain began to produce H2O2 during the exponential phase and reached a maximal concentration of 150 mg/L on the stationary phase, with a maximal OD600nm of 0.3. This concentration was five times higher than those observed for Lactococcus garvieae strain. The same H2O2 production profile was observed for C. inhibens CD344, DSM 13024, and WN1359 strains (data not shown).  Figure 6. Assessment of H 2 O 2 production during the growth of Carnobacteriun inhibens MIP2551 (triangle) and Lactococcus garvieae CIP102507 (circle). Dashed lines: H 2 O 2 concentration estimated by Dosatest ® Peroxide test strips; Solid lines: OD measured at 600 nm. Strains were cultivated at 26 • C, in brain heart infusion (BHI) medium and in shaking conditions.

Comparison of CFSs Activities
Sixty-five strains (56 C. maltaromaticum and nine C. divergens) remained active after catalase treatment (Table S3). Their cell-free supernatants (CFSs) were tested against Lactococcus garvieae and L. monocytogenes. All the 65 CFSs found active against L. monocytogenes remained active, suggesting an extracellular active compound. However, only three CFSs remained active against Lactococcus garvieae. C. maltaromaticum SF668 was the only strain with a CFS active against both bacterial targets. After digestion by proteinase K, all CFSs became inactive, suggesting a peptidic nature for antimicrobial compounds, such as bacteriocins.
CFS efficacy against L. monocytogenes was determined and compared for all active strains. The growth of L. monocytogenes was measured in CFS-supplemented medium. The 65 active CFSs were diluted to determine the minimal inhibitory dilution (MID) (Figure 7). The MID varied from less than 2 to 256 for C. divergens, and less than 2 to 2048 for C. maltaromaticum strains. Seventy one percent C. maltaromaticum CFSs (40 out of 56) and 22% C. divergens CFSs (two out of nine) showed a MID ≥ 256. This result suggested higher activity for C. maltaromaticum strains. The C. divergens V41, which produces the antilisterial divercin V41 (Table S1) displayed a MID = 16 in the tested conditions. Three C. divergens strains, namely CD317, CD320, and CD349, showed higher activity than the C. divergens V41 strain. C. maltaromaticum V1 strain, which produces the previously described carnobacteriocin BM1 and piscicolin 126 [64] (Table S1), and the EBP3019 and SF668 strains belonged to the most efficient CFS group. A MID greater than 512 was only detected in the EBP3019 strain. This result was confirmed by supplementary triplicate assays allowing the estimation of an average MID of 1024.

MIP2551, EBP3019, and SF668 Genome Specificities
It appeared from all these analyses that C. maltaromaticum EBP3019 and SF668 strains displayed interesting antimicrobial activities related to extracellular peptides. SF668 strain was previously shown to have antilisterial effect in cold-smoked salmon and used as bioprotective strain [25].
In addition, C. inhibens MIP2551 was shown to produce high amounts of H 2 O 2 . In order to identify potentially original antimicrobial BGCs, the genomes of these three strains were sequenced, assembled, and annotated. Phylogenomic analysis by ANI similarity values for these three strains confirmed the identification using partial 16S rDNA sequencing (Figure 1).
Genes coding for H 2 O 2 anabolism and catabolism were identified in the MIP2551 genome and compared to other species (Figure 8). In most cases, it seemed to be species dependent. The catalase gene involved in H 2 O 2 catabolism was found in MIP2551, and in 88% of the analyzed genomes. Genes coding for the oxidases GplO, Pox, and Lox were found in C. inhibens MIP2551 genome. This genetic content was similar to environmental species, which is consistent with H 2 O 2 inhibition in our bioassay (excepted for C. antarcticum not represented). Two to three genes coding for Lox were detected in some genomes as in MIP2551.  compared to other species (Figure 8). In most cases, it seemed to be species dependent. The catalase gene involved in H2O2 catabolism was found in MIP2551, and in 88% of the analyzed genomes. Genes coding for the oxidases GplO, Pox, and Lox were found in C. inhibens MIP2551 genome. This genetic content was similar to environmental species, which is consistent with H2O2 inhibition in our bioassay (excepted for C. antarcticum not represented). Two to three genes coding for Lox were detected in some genomes as in MIP2551.  The BGC content of the C. inhibens MIP2551, C. maltaromaticum EBP3019, and SF668 strains was investigated, and the predicted clusters were dereplicated. The two undescribed terpene BGCs T1 and T3 were predicted in C. inhibens MIP2551 genome. A total of five RiPPs were predicted in the SF668 strain, identified as carnolysin A1/A2, piscicolin 126, carnobacteriocin BM1, B2, and X/Y (Figure 2). This equipment corresponded to the largest RIPPs BGC diversity among the Carnobacterium spp. genomes. Three RiPPs BGCs were predicted for the C. maltaromaticum EBP3019 strain-the highly conserved carnobacteriocin BM1 and the maltaricin CPN BGC [26] along with a new UB named UB5 (Figures 2 and 3). This last BGC displayed the best sequence similarity (49%) with propionicin SM1 from Propionibacterium jensenii [65] (Table 1). UB5 is probably extracellular and has a high molecular weight (16.86 kDa).

Discussion
The use of phylogenomics and, in particular, the analysis of ANI allows new insights into genome evolution and bacterial species definition [66,67]. Within the genus Carnobacterium, only the species C. alterfunditum and C. pleistocenium cannot be distinguished on the basis of 16S rDNA alone. In this study, a combination of 16S rDNA-based phylogeny and ANI similarity ensured accurate strain identification.
The constitution of a reliable Carnobacterium spp. genome dataset allowed us to discriminate BGC content depending on species and ecological niches. The number of BGCs, and the class of the predicted molecules appeared to be generally species-specific. For example, carnolysin A1/A2 and carnobacteriocin BM1 are specifically found in C. maltaromaticum and thiopeptide Th1 in C. divergens strains. The food-sourced species C. maltaromaticum appeared to be particularly rich in RiPPs, as well as C. divergens to a lesser extent. These observations could be related to their adaptation to animal and food product habitats as previously described [20]. In the same way, RiPPs with original sequences, were identified in strains belonging to the environmental species C. jeotgali, C. inhibens, and C. viridans but isolated in food products ( Figure 2). The acquisition of RiPPs in these environmental species could be due to the adaptation to fight against Gram-positive strains largely present in the food microbiome and absent in oceans. This is consistent with a previous study showing that the MIP2551 strain lacking in RiPPs was not competitive in salmon gravlax [68]. Other BGC such as terpene were exclusively predicted in environmental species, with the exception of BGC T2. Terpenes are less studied in bacteria than in plants or fungi, but they do represent a promising source of bioactive compounds. This class of BGCs has not been described yet in the genus Carnobacterium. More genome sequences for each species are needed to confirm these observations.
In this study, C. divergens and C. maltaromaticum isolated from seafood products represented more than 95% of the isolated strains, a ratio which is consistent with that observed in the metagenomic analyses of food [20]. No activity related to the acidification or the production of organic acids was detected in our experiments, which reinforces data from previous studies [69,70]. Genomic analyses revealed the presence of BGCs in most of the C. maltaromaticum and C. divergens strains while only 60% of the tested isolates were active. Thanks to the reference strains screened for antimicrobial activity, we also observed that some of them showed no activity despite the presence of terpene, NRPS and RiPPs in their genome (e.g., C. mobile DSM4848, C. inhibens DSM13024, C. viridans MPL-11). Since these BGCs remain undescribed, it could be hypothesized that they encode molecules devoid of antimicrobial activity, or active against other strains not tested in this screening. BGCs might also not be expressed under the culture conditions used. Further investigations using culturomic analyses or co-cultures could help to activate such cryptic BGCs [71,72].
The antimicrobial screening showed that environmental Carnobacterium species such as C. inhibens, C. jeotgali, C. viridans more likely inhibit bacterial growth through H 2 O 2 accumulation. High H 2 O 2 level was in accordance with genomic differentiation of niche species, particularly the presence of lox and pox genes compared to C. maltaromaticum. Most C. divergens strains also produced H 2 O 2 without harboring the known oxidase encoding genes lox, pox, and glpO ( Figure 8). Lox gene encodes the lactate oxidase responsible for the conversion of lactic acid into pyruvate. It can then be hypothesized that the presence of both lox and pox genes gives a double advantage to outcompete other microorganisms not only through nutritional competition but also through the production of antimicrobial compounds. C. inhibens, C. jeotgali, and C. viridans species harbor multiple copies of genes encoding Lox probably correlated with H 2 O 2 production [73], and in parallel the T3 BGC similar to staphyloxanthin-encoding BGC, which has protective properties against H 2 O 2 . A staphyloxanthin-similar function can, therefore, be assumed for the T3 cluster, although no xantho-pigment was produced by the T3 cluster containing strains.

Conclusions
Combining genome mining, phenotype characterization, and dereplication on a large dataset appeared a relevant approach to avoid rediscovering known active molecules and their BGCs. Thanks to this strategy, new insights into the antimicrobial potential of environmental Carnobacterium species suggests that they should be considered for further new biotechnological applications. Furthermore, BGC distribution in Carnobacterium spp. showed that the screening of active RiPPs should be performed in food-related strains. Two of them, EBP3019 and SF668, were selected for their activity against Listeria monocytogenes. Further studies on EBP3019 supernatant, such as fractionating, purification, mass analysis, and amino acid sequencing have to be conducted to isolate and characterize UB5, whose BGC was not found in any other genome.