Lactococcus lactis Diversity Revealed by Targeted Amplicon Sequencing of purR Gene, Metabolic Comparisons and Antimicrobial Properties in an Undefined Mixed Starter Culture Used for Soft-Cheese Manufacture

The undefined mixed starter culture (UMSC) is used in the manufacture of cheeses. Deciphering UMSC microbial diversity is important to optimize industrial processes. The UMSC was studied using culture-dependent and culture-independent based methods. MALDI-TOF MS enabled identification of species primarily from the Lactococcus genus. Comparisons of carbohydrate metabolism profiles allowed to discriminate five phenotypes of Lactococcus (n = 26/1616). The 16S sequences analysis (V1–V3, V3–V4 regions) clustered the UMSC microbial diversity into two Lactococcus operational taxonomic units (OTUs). These clustering results were improved with the DADA2 algorithm on the housekeeping purR sequences. Five L. lactis variants were detected among the UMSC. The whole-genome sequencing of six isolates allowed for the identification of the lactis subspecies using Illumina® (n = 5) and Pacbio® (n = 1) technologies. Kegg analysis confirmed the L. lactis species-specific niche adaptations and highlighted a progressive gene pseudogenization. Then, agar spot tests and agar well diffusion assays were used to assess UMSC antimicrobial activities. Of note, isolate supernatants (n = 34/1616) were shown to inhibit the growth of Salmonella ser. Typhimurium CIP 104115, Lactobacillus sakei CIP 104494, Staphylococcus aureus DSMZ 13661, Enterococcus faecalis CIP103015 and Listeria innocua CIP 80.11. Collectively, these results provide insightful information about UMSC L. lactis diversity and revealed a potential application as a bio-protective starter culture.


Introduction
Artisanal soft cheeses are a class of handmade products following ancient practices and manufactured with a limited mechanization process. These cheeses include the use of undefined mixed starter cultures (UMSC) for traditional dairy-based food fermentations [1][2][3]. UMSC are composed of undefined multiple-strain cultures of lactic acid bacteria (LAB). They harbor complex microbiota, which naturally proceed from raw milk, and are specific to the producing region [4].
Most important bacteria in dairy UMSC are Lactococcus lactis, Leuconostoc mesenteroides, Streptococcus spp. and Lactobacillus spp. [1,3,5,6]. However, limited knowledge is currently available on the genetic diversity of UMSC [7]. Changes caused by intrinsic milk properties, but also by other factors such as the dairy environment and cheese making procedures, can influence the UMSC microbial composition. Therefore its complexity is specific beyond the species/subspecies level [8][9][10]. From the beginning of dairying, Lactococci have been widely used as a principal component of starter cultures [6,11]. Both subspecies L. lactis subsp. lactis and L. lactis subsp. cremoris are the upmost contributors to the acidification process following the rapid fermentation of lactose [7]. The diversity of Lactococcus strains provides an important reservoir for starters and may contribute to flavor differences and specific features in a fermented milk-product [12]. Lactococci participate in food preservation, texture formation and flavor by producing aromatic compounds (alcohols, aldehydes, ketones) or by metabolizing substrates such as citrate, amino acid or fats [13]. The growing interest to preserve traditional cheeses has stimulated the use of raw milk UMSC and thus, a better understanding of the diversity of Lactococcus lactis strains is required. High-throughput sequencing (HTS) and whole-genome sequencing (WGS) technologies have opened a new window to decipher such microbial diversity. Lately, industrial and technological relevance has motivated extensive genomic studies [14][15][16] to better predict production performance and advise culture strategies to avoid fermentation failures [17]. Comparative genomics has also provided insightful information into genotype-phenotype associations. Phenotypic data and Lactococci strain-specific information are effective methods for the assignment of functions underlying industrial traits or niche adaptation [18,19].
Starter culture use implies the respect of safety and quality patterns in the dairy industry. Spoilage bacteria and foodborne pathogens are of major concern in food production and processing environments as well as for consumers since they cause economic losses and foodborne illnesses. Some LAB species, especially L. lactis, can exert a food preservative effect due to a nutritional competition ability or production of organic acids or antimicrobial compounds. The pH reduction makes the dairy matrix unfavorable for the growth of food-spoilage or pathogenic microorganisms. LAB can also produce metabolic end-products including hydrogen peroxide (H 2 O 2 ), CO 2 , diacetyl, acetaldehydes, D-isomers of amino acids, reuterin and bacteriocins [20,21]. These antimicrobial strategies are increasingly used for limiting the growth of unwanted microorganisms [22]. In a raw milk starter culture, the food micro-structure is in a liquid form and favors the accumulation of biomass and metabolites causing chemical changes such as a decline of pH. In a gel food microstructure (such as in cheese), microorganisms are immobilized and constrained to form colonies accompanied by local accumulation of metabolic end-products. The immobilization of microorganisms in a cheese sample might result in interactions between colonies and local pH changes to stimulate an acid tolerance response. Therefore, food preservation depends also on the structure of the product [23].
In this study, genotype-phenotype associations and bacterial screenings were carried out to characterize the UMSC bacterial diversity to better understand UMSC carbohydrate metabolism and to evaluate antimicrobial activities.

Sample Collection
The UMSC was obtained following raw milk treatment as described in the Bionatif ® patented process [24]. The raw milk was obtained from a farm located in Normandy (Isigny-sur-mer, France), and the UMSC was used daily for soft-cheese manufacture. The UMSC was kept frozen at −80 • C (Flores de terroirs, Aytré, France).

DNA Extraction
Samples of UMSC were aseptically homogenized and centrifuged at 13,000× g for 10 min at 4 • C. DNA was extracted with the DNeasy PowerFood Microbial Kit (Qiagen, Hilden, Germany), which combines mechanical (beat beating) and chemical lysis of the cell. The purified DNA was stored at −80 • C.

Gene Sequencing and Data Analysis of 16S rRNA
The V1-V3 and the V3-V4 regions of the 16S rRNA genes were amplified by PCR. The V1-V3 bacterial universal primers were selected [25,26]. Primers 27F (3 -AGAGTTTGATCATGGCTCAG-5 ) and 1492R (3 -TACGGYTACCTTGTTACGACTT-5 ) were used for the V3-V4 amplification fragment. For each sample, an amplified DNA quality check was performed. The amplified DNA was then sequenced by the Eurofins Genomics platform (Ebersberg, Germany) using an Illumina ® MiSeq Technology.
The FROGS (find rapidly OTUs with Galaxy solution) pipeline (version 3.1) was used to retrieve the operational taxonomic units (OTUs) under the Galaxy environment [27]. This pipeline includes raw FASTQ files demultiplexing and a preprocess step to filter and delete sequences with unexpected lengths, with ambiguous bases (N) and the ones that do not have primer sequences at both 3 and 5 ends. The clustering of sequences was performed by Swarm using an iterative growth process with a local clustering threshold of 3 [28]. Chimera sequences were identified and removed using Vsearch [29]. OTUs accounting for more than 0.5% of the total reads were considered. The reference sequences of each OTU were blasted against the SILVA database (version 128) for taxonomic assignation [30].
The data were analyzed using the DADA2 pipeline (v. 1.12, http://benjjneb.github.io/dada2/ tutorial.html), which retrieves biological sequences from reads by modelling the Illumina-sequencing errors [31]. Once read quality profiles were validated, the core-denoising algorithm of DADA2 was performed on the forward and reverse reads separately. Amplicon sequence variants (ASV) were inferred and pairs merged. ASVs table was constructed and chimeras were identified and removed only if the parent sequence was at least twice abundant than a left/right-segment recombination. The taxonomic affiliation was assigned using a local BLAST database produced using the lactococcal genomes available on the NCBI database.

Bacterial Library
Presumptive LAB were enumerated on MRS, BHI or M17 agar (Biokar, Allonne, France) according to the conditions reported in Table 1. Bacterial isolates were isolated (n = 1616), purified twice and maintained frozen at −80 • C in the corresponding growth medium containing 20% (v/v) glycerol (Table 1). Six L. lactis isolates were chosen following API 50 CH results to perform genotype-phenotype associations. Overnight cultures from these isolates were centrifuged at 13,000× g for 10 min at 4 • C. These. Genomic DNA was extracted using the DNeasy blood and tissue kit (Qiagen, Germany) according to the manufacturer's instruction. Genomic DNA was quantified using the NanoDrop ND-1000 (ThermoFisher Scientific, Illkirch-Graffenstaden, France).

Whole Genome Sequencing and Gene Annotation
Whole-genome sequence data were generated by PacBio ® RS II and Illumina ® iSeq sequencing technology in order to produce de novo assemblies. The PacBio ® RS II data were obtained and assembled using the Hierarchical Genome Assembly Process (HGAP) assembly algorithm [34], on the Eurofins Genomics platform (Ebersberg, Germany). The Illumina ® iSeq data were produced and assembled using the SPAdes algorithm (Mérieux Nutrisciences, Nantes, France) [35]. Chromosomal organization of the resulting contigs were predicted by aligning them against the L. lactis subsp. lactis IL1403 reference genome using the MicroScope platform [36]. The individual genome assemblies were deposited in the NCBI database with accession codes listed in Table 2. Overall genomic relatedness between genomes was assessed using the Orthologous Average Nucleotide Identity (OrthoANI) tool [37]. Carbohydrate metabolism-related pathway information was identified from those described in the Kyoto Encyclopedia of Genes and Genomes (KEGG, http://www.kegg.jp/) [38]. The mining of genes clusters related to the production of antimicrobial peptides was performed using BAGEL4 [39].

Accession Number(s)
The 16S and purR amplicon data were deposited at DDBJ/ENA/GenBank under BioProject number PRJNA615620. The whole-genome project was deposited at DDBJ/ENA/GenBank under BioProject number PRJNA615395.

Bacterial Strains and Growth Conditions
Indicator strains were selected from culture collections and maintained at −80 • C in their growth medium supplemented with 20% glycerol (v/v). They were cultured twice in growth conditions specific to each strain (Table 3).  An agar spot test was used to determine the antimicrobial potential of LAB isolates against the indicator strains (Table 3). Five microliters of an overnight culture (approximately 10 8 CFU/mL) of each LAB isolate were spotted onto BHI agar plates (15 g/L agar) and left to dry for 20 min at room temperature under sterile conditions. The plates were then overlaid with 10 mL of soft agar (8 g/L) seeded with 10 6 CFU/mL of the indicator strain (Table 3). Plates were incubated overnight at the optimum growth temperature of each indicator strain and checked for inhibition halos. As positive control, Lactobacillus sakei CIP 104,494 was used because of its large antimicrobial spectrum [41].

Preparations of Cell-Free Supernatant (CFS) and Neutralized CFS (NCFS)
Cell-free supernatants (CFSs) were obtained by centrifugation at 8600 g for 10 min at 4 • C to separate bacterial cells and supernatants. The neutralized CFS (NCFS) were prepared by adjusting the pH of CFS to 7.0 with NaOH 1M to exclude the antimicrobial effect of organic acids. Samples were heated at 100 • C for 10 min to inhibit the enzyme activity. The antimicrobial properties of CFSs and NCFSs were tested on the indicator strains by the agar well diffusion assay.
Agar Well Diffusion Assay BHI agar was seeded with 10 6 CFU/mL of indicator strains (Table 3), homogenized and poured into sterile Petri dishes. After drying under sterile conditions, sterile glass wells were placed on agar plates. Aliquots of non-diluted CFS or NCFS produced by LAB isolates were used to fill in the wells (100 µL). Plates were incubated overnight at the optimum growth temperature of each indicator strain. Inhibition halos were observed and recorded.

Inhibitory Effect during Co-Cultivation
The ability of LAB isolates to inhibit the growth of S. aureus DSMZ 13,661 and S. ser. Typhimurium CIP 104,115 was further studied during co-cultivation assays in a double concentrated BHI media (pH 7) [42]. The co-cultivation process consisted of two different culture batches. Batch 1 corresponded to the co-inoculation of S. aureus DSMZ 13,661 with the selected LAB isolate and batch 2, the co-inoculation of S. ser. Typhimurium CIP 104,115 with the LAB isolate. Growth control was performed using a single culture for each indicator strain or the LAB isolate. An inoculum of 1% (v/v) of LAB isolate overnight culture (10 9 CFU/mL in BHI broth) and 1% (v/v) of indicator strain overnight culture (10 9 CFU/mL in BHI broth) were seeded to perform Batch 1 or Batch 2 co-culture assay. All batches were incubated at 28 • C for 48 h. The CFS from each batch culture was collected every 2 h. Antimicrobial activities were detected by the agar well diffusion assay as described above.

UMSC Bacterial Diversity Studies
Cheese manufacture behavior, processing environment and starter cultures are known to harbor an indigenous microbiota that influences organoleptic features on artisanal cheeses. The bacterial composition of starter cultures supports organoleptic and biochemical changes during ripening [43]. Revealing information on microbial consortia are essential for artisanal cheesemakers, considering the willingness of the consumer market to require locally manufactured products with different organoleptic properties [9,43,44]. Part of a daily routine, the studied UMSC is used for soft-cheese manufacture. Its microbial transfer is applied during rennet addition, a few hours after milk enrichment with acidifying starters. This cheese-making practice, combined with a high diversity of microbial activities, is key to developing gustatory characteristics. In this study, the microbial consortium of the UMSC was deciphered using culture-independent and culture-dependent based methods.
First, the UMSC microbial diversity was assessed using three different culture-independent methods: (i) analysis of 16S rRNA gene fragment sequences, (ii) purR gene sequence analysis, (iii) whole-genome sequencing analysis. Afterward, two culture-based methods were used on bacterial isolates recovered from the UMSC: MALDI-TOF MS and full-length 16S rRNA gene sequencing analysis.

Analysis of 16S rRNA Gene Fragment Sequences
An assessment of the microbial diversity started with targeted amplicon sequencing of 16S rRNA gene fragments. Read counts (total of 65,447 reads for V3-V4 and 61,383 reads for V1-V3) were filtered and relative abundances were calculated for each OTU. One dominant genus, Lactococcus sp., encompassing 99.9% of relative abundance was identified for each 16S rRNA gene region. With a clustering threshold of three, two OTUs were obtained among the Lactococcus genus.

purR Gene Sequence Analysis
The housekeeping gene purR was then studied to complete the assessment of microbial diversity. It was previously selected after pan-genome analysis for the differentiation of clades beyond the L. lactis subspecies level [7]. The purR gene encodes for a transcriptional regulator of purine biosynthesis. The use of a single nucleotide inference method highlighted five amplicon sequence variants (ASVs) after raw data processing using the DADA2 algorithm (Table 4). A total of 108,455 reads were obtained with an exact purR amplicon size of 435 nucleotides for each clustered ASV [45].  Figure 1A). The nucleotide translation into amino acid sequences identified three different ASVs (see matrix table, Figure 1B). Amino acids sequences of variants ASV_1, ASV_2 and ASV_3 were identical. Variants ASV_4 and ASV_5 showed three amino acid substitutions compared to other ASV (or the reference sequence). A comparison between ASV_4 and ASV_5 showed one amino acid substitution. Substitutions include amino acid changes in position 7: Glu (ASV_1), Ala (ASV_4) and Ser (ASV_5), position 21: Thr (ASV_1) and Pro (ASV_4 and ASV_5) and position 22: Lys (ASV_1) and Gln (ASV_4 and ASV_5). Dendrograms were constructed using nucleotide alignments ( Figure 1A) or protein alignments ( Figure 1B). . purR genes were sequenced from the undefined mixed starter culture (UMSC) global genomic DNA extraction. The tree was constructed using Phylogeny.fr [46]. Matrix tables highlight differences in the composition of nucleotide (A) or amino acid (B) sequences.
These five ASVs sequence reads were all affiliated to the species level (100% identity) as L. lactis using BLAST searches against a local database including any sequences of lactococcal genomes part of the NCBI database (built on 10 February 2020). Interestingly, the phylogenetic analysis of purR did not identify L. lactis to the subspecies level, as it did for Frantzen et al. [7]. In their work, primers designed by Frantzen et al. were used for relative quantification of the microbial community in starter cultures. Medium amplicon size was then required. Therefore, the purR amplicon size (435 nucleotides) was shorter than the full-length purR gene (816 nucleotides), identified in the wholegenome sequences. The targeted amplicon purR showed an alignment between position 347 to 782. A BLAST search performed on the full-length purR gene affiliated the genome sequences to the subspecies level (100% identity) as L. lactis subsp. lactis. These results suggested that new primers devised to amplify longer fragments of the purR gene would be needed to explore L. lactis diversity to the subspecies level in the UMSC. Another approach was attempted using the eps operon, both located on plasmids and chromosomes [7]. Unfortunately, irregular amplicon sizes were obtained, and it was impossible to differentiate isolates to the species/subspecies level (data not shown).
The ASV method infers the biological sequences in the sample before the introduction of amplification and sequencing errors. It distinguishes sequence variants differing by as little as one nucleotide [47]. The OTU assignment methods gather clusters of reads that differ by less than a fixed clustering threshold of three [28]. In a previous study, Lactococci strains were confirmed to have multiple copies of the 16S rRNA gene. These copies harbor intra-genomic heterogeneity, which makes it difficult to differentiate bacteria to the species/subspecies level using 16S rRNA amplicon [45,48]. ASV methods better discriminate ecological patterns capturing all biological variation [31]. Furthermore, DNA sequences of protein-coding genes are more effective than 16S rRNA genes when distinguishing between very closely related bacteria [7,49]. The targeted amplicon comparison highlighted a noteworthy difference between results obtained with fragments of the 16S rRNA gene . purR genes were sequenced from the undefined mixed starter culture (UMSC) global genomic DNA extraction. The tree was constructed using Phylogeny.fr [46]. Matrix tables highlight differences in the composition of nucleotide (A) or amino acid (B) sequences.
These five ASVs sequence reads were all affiliated to the species level (100% identity) as L. lactis using BLAST searches against a local database including any sequences of lactococcal genomes part of the NCBI database (built on 10 February 2020). Interestingly, the phylogenetic analysis of purR did not identify L. lactis to the subspecies level, as it did for Frantzen et al. [7]. In their work, primers designed by Frantzen et al. were used for relative quantification of the microbial community in starter cultures. Medium amplicon size was then required. Therefore, the purR amplicon size (435 nucleotides) was shorter than the full-length purR gene (816 nucleotides), identified in the whole-genome sequences. The targeted amplicon purR showed an alignment between position 347 to 782. A BLAST search performed on the full-length purR gene affiliated the genome sequences to the subspecies level (100% identity) as L. lactis subsp. lactis. These results suggested that new primers devised to amplify longer fragments of the purR gene would be needed to explore L. lactis diversity to the subspecies level in the UMSC. Another approach was attempted using the eps operon, both located on plasmids and chromosomes [7]. Unfortunately, irregular amplicon sizes were obtained, and it was impossible to differentiate isolates to the species/subspecies level (data not shown).
The ASV method infers the biological sequences in the sample before the introduction of amplification and sequencing errors. It distinguishes sequence variants differing by as little as one nucleotide [47]. The OTU assignment methods gather clusters of reads that differ by less than a fixed clustering threshold of three [28]. In a previous study, Lactococci strains were confirmed to have multiple copies of the 16S rRNA gene. These copies harbor intra-genomic heterogeneity, which makes it difficult to differentiate bacteria to the species/subspecies level using 16S rRNA amplicon [45,48]. ASV methods better discriminate ecological patterns capturing all biological variation [31]. Furthermore, DNA sequences of protein-coding genes are more effective than 16S rRNA genes when distinguishing between very closely related bacteria [7,49]. The targeted amplicon comparison highlighted a noteworthy difference between results obtained with fragments of the 16S rRNA gene (2 OTUs) or with the purR gene (5 ASVs). The housekeeping purR gene seems to be the best candidate since it enabled the differentiation of clades beyond the species level.

Whole-Genome Sequencing Analysis
The ANI value is used to estimate the genetic distance between isolates, which reflects the sequence identities of the conserved regions between two genomes [50]. Whole-genome sequencing and assemblies were performed. The pairwise OrthoANI values of six in-house Lactococcus and one reference genome (L. Lactis subsp. lactis IL1403) were calculated to assess their genetic relatedness. One cluster was retrieved (Supplementary Materials Figure S2). Based on the guidelines set out by Richter and Rosselló-Móra, L. lactis subsp. lactis strains are a different species to subspecies cremoris strains possessing an ANI value below 95% [51]. Here, the ANI identity was >99.8% for all of the L. lactis genomes, suggesting they all belong to the lactis subspecies group.

Bacterial Library
Second, a culture-dependent based method was performed to complete the study of the UMSC bacterial diversity. A bacterial library was established from the UMSC with a set of 1616 bacterial isolates, according to 13 growth conditions (Table 5).

MALDI-TOF MS and Full-Length 16S rRNA Gene Sequencing Analysis
Two representative isolates of each growth condition were randomly selected for MALDI-TOF MS and full-length 16S rRNA gene sequencing characterizations. A set of 26 bacterial isolates (n = 26/1616) was analyzed as a sample that could represent the bacterial diversity of each growth condition. Four species identifications were obtained: Lactococcus lactis (L) (n = 18/26), Lactococcus raffinolactis (R) (n = 1/26), Enterococcus faecalis (E) (n = 5/26) and Staphylococcus warneri (S) (n = 2/26). The culture-dependent analysis showed similarities in the UMSC diversity and agreed with culture-independent studies for the L. lactis identification. Results also highlighted the presence of other species besides L. lactis. These latter could be part of the remaining 0.1% of relative abundance outlined by 16S rRNA targeted amplicon sequencing. In terms of genus richness and effective number of genera, the differences observed among 16S rRNA analytical method were most likely a result of unclassified reads in the bioinformatic pipeline [52].

UMSC Genotype-Phenotype Associations
API 50 CH systems were used to explore carbohydrates metabolism. Height carbohydrate metabolism profiles were retrieved from the previous 26 examined LAB isolates (Supplementary Materials Figure S1). Three dairy L. lactis type strains were also tested ( Table 6). Carbohydrate metabolism plays a key role in the growth of Lactococcus species. On average, 12.5% of genes are associated with carbohydrate metabolism and involved in glycolysis/gluconeogenesis and the pentose phosphate pathway (PPP) [53]. Four L. lactis phenotypic groups were obtained: (A) for isolates L1 to L6, L10 to L13 and L15 to L17 (n = 13); (B) for isolates L7 and L8 (n = 2); (C) for isolates L14 and L18 (n = 2) and (D) for isolate L9 (n = 1) ( Table 6).  [14,53]. Phenotypic data of the type strain IL1403 was extracted from Siezen et al. [54].  A genotype-phenotype association was devised to study possible relations between genes and phenotypes observed in the API 50 CH systems. A representative set of six L. lactis isolates were selected for WGS studies: three isolates were part of the A group (L1, L2, L3), one isolate was part of the B group (L8), one isolate was part of the C group (L14) and one isolate was part of the D group (L9). An Illumina iSeq ® platform (L1, L2, L8, L9, L14 genomes) and a PacBio ® RS II technology (L3 genome) were used for genome assemblies. The role of L. lactis as an important industrial starter strain is mainly due to the rapid conversion of lactose to lactic acid. The KEGG pathways (https://www.genome.jp/kegg-bin) of Lactococci cell metabolism summarize two possible catabolic ways for lactose utilization: (i) hydrolysis of lactose to give α-D-glucose and D-galactose performed by a β-galactosidase (EC 3.2.1.23), and (ii) the conversion into a lactose 6-phosphate (Lac-6P) by a lactose phosphotransferase (EC 2.7.1.207). The Lac-6P is subsequently hydrolyzed by 6-phospho-β-galactosidase (EC 3.2.1.85) to give D-galactose-6P (Gal-6P), and then Gal-6P is catabolized by the tagatose 6-phosphate pathway [55]. Genome sequence analysis highlighted both catabolic pathways with genes encoding these key enzymes. Conversely, phenotypic results indicated that the B group was able to degrade D-sorbitol, although it lacked all gene families found to be significantly associated with this degradation. Buron-Moles et al. reported the same results with dairy L. lactis isolates [14].
Regarding D-tagatose degradation, phenotypic results observed in the API 50 CH systems revealed that isolates of the B group could use this carbohydrate source whilst other isolates (or type strains) could not. D-tagatose is hydrolyzed by a tagatose-6-phosphate kinase (EC 2.7.1.144) to give D-tagatose-6P (Tag-6P). Tag-6P is subsequently hydrolyzed by a tagatose-1,6-diphosphate aldolase (EC 4.1.2.40) to give Tag-6P 2 , and then it enters in the PPP. These enzymes were found in each studied genome even though the B group was the only one able to grow on D-tagatose. By studying genotype-phenotype associations between gene families and their carbohydrate metabolism, Buron-Moles et al. reported that D-tagatose can only be degraded by a few LAB members of Lactobacillus and Pediococcus strains [14]. Siezen et al. also reported the inability of Lactococcus strains to grow on D-tagatose [54]. Here, the L. lactis isolates were highlighted for their uncommon D-tagatose degradation.
D-xylose units are components of the β-1,4-xylan, a hemicellulosic polysaccharide of plant materials [54]. The β-1,4-xylosidase (EC 3.2.1.37) is a key enzyme involved in the β-1,4-xylan degradation to obtain D-xylose oligomers. D-xylose is then converted by a xylose isomerase (EC 5.3.1.5) to form D-xylulose (D-xyl). D-xyl is then catabolized by a xylulokinase (EC 2.7.1.17) to enter in the PPP. Genes encoding these enzymes were found in each analyzed genome. The carbohydrate metabolism study demonstrated that isolates were able to use D-xylose, except for those that were part of the B group, the cremoris subspecies type strain and the dairy strain L. lactis IL1403. Similarly, phenotypic results highlighted isolates able to grow on D-sucrose except for group A and L. lactis type strains, which were sucrose-negative. A previous study also reported the inability of dairy Lactococcus strains to utilize D-sucrose [14], unlike strains from plant origin that could grow on the D-sucrose [18,54]. Genotypic relationships showed no genes encoding for D-sucrose degradation. As noted by Buron-Moles et al., these findings emphasize that phenotypes cannot be always inferred from the presence/absence of specific genes but require the explicit integration of phenotypic information [14].
Phenotypic features in carbohydrate metabolism appeared to reflect the functional diversity observed for each L. lactis isolate. Moreover, the percentage of carbohydrates utilized per isolate was slightly higher (33% to 45%) in comparison with L. lactis type strains (29%). Their metabolism potential exhibited differences and seemed to be richer and diversified. Interestingly, the presence of some genes did not always imply a degradation of the associated carbohydrate in the API 50 CH systems. It could be explained by a progressive pseudogenization process (gene inactivation) linked to their current niche specification. The L. lactis isolates came from a farm environment before being introduced into raw milk and then into the UMSC. A comparative study revealed that an L. lactis gene set related to degradation products such as arabinose and xylose originated from plants [54]. Genomes of subspecies lactis KF147 and KF282, which were of plant origin, possessed genes set for the uptake and conversion of typical cell wall degradation products such as arabinose and xylose. All of the analyzed genomes possessed a key enzyme for L-arabinose degradation even though phenotypic profiles showed an inability to grow on arabinose (Table 6). Moreover, B group isolates possessed the araD gene, encoding an L-ribulose-5-phosphate 4-epimerase (EC 5.1.3.4), but it was annotated as a pseudogene. A mutation may have divided the gene into two-parts, and thus the araD gene became inactive. These results suggested that the human-made environment triggered microbial strains to genomic function adaptations to their dairy environment. UMSC isolates progressively appeared to be domesticated since they no longer could grow in a variety of plant-derived carbohydrates due to pseudogenization or transposon insertions in genes [56]. These genomic changes could be worsened due to the "backslopping" process in which artisans are continuously transferring microbial material from the previous fermentation product to start a new batch [15].

Antimicrobial Activities
The antimicrobial activity of bacterial isolates against foodborne, food-spoilage and pathogenic bacteria was evaluated by the agar well diffusion assay. All isolates that were part of the bacterial library (n = 1616) were tested. A total of 34/1616 isolates showed antimicrobial activity against up to five indicator strains (Table 7 and Supplementary Materials Table S1). Isolates 14 and 15 showed the largest antimicrobial spectrum inhibiting the growth of the five indicator strains (S. ser. Typhimurium CIP 104115, S. aureus DSMZ 13661, S. aureus CIP 76.25, E. faecalis CIP 103,015 and L. innocua CIP 80.11). S. aureus DSMZ 13,661 was sensitive and showed the largest inhibition zones. S. ser. Typhimurium CIP 104,115 was the only Gram-negative bacteria to be sensitive toward isolates 14 and 15. The antimicrobial activity of the L. lactis-isolate 14 was further studied in co-cultivation assays against the two indicator strains: S. aureus DSMZ 13,661 or S. ser. Typhimurium CIP 104115.
During co-culture assays, the growth of each bacterium was monitored by bacterial counts on a specific medium (data not shown). Growth controls were obtained using bacterial counts performed on single cultures of each strain (indicators and isolate 14). Temporary growth slowdowns were observed during co-cultures and compared with their respective growth controls. Growth slowdowns in co-culture assays were highlighted in the exponential growth phase of S. ser. Typhimurium (between 8 and 20 h) and in the stationary growth phase of S. aureus DSMZ 13,661 (between 26 and 34 h). Their respective controls showed no growth disruption. The growth in the co-culture of isolate 14 remained similar to that of the growth control containing no indicator strain (data not shown).
To further examine isolate 14 antimicrobial activity, CFSs of each batch culture were collected during these growth slowdowns. Their antimicrobial activity was assessed by the agar well diffusion assay. Thus, CFSs were collected between 26 and 34 h of growth for S. aureus DSMZ 13,661 assays (mono and co-cultures). Their CFSs largest inhibition halo was obtained after 32 h of growth. Similarly, CFSs were collected between 8 and 20 h of growth for S. ser. Typhimurium CIP 104,115 assays (mono and co-cultures). Their CFSs largest inhibition halo was obtained after 16 h of growth (data not shown). Table 8 summarizes inhibition halo values (mm) of CFSs both obtained from mono and co-cultures after 32 h of growth for S. aureus assays and after 16 h of growth for S. ser. Typhimurium assays. CFSs of isolate 14, alone or in the presence of indicator strains, showed antimicrobial activity against S. ser. Typhimurium CIP 104115, Lactobacillus sakei CIP 104494, S. aureus DSMZ 13661, L. innocua CIP 80.11 and L. lactis subsp. lactis IL1403. CFSs of indicator strains showed no inhibition halo against themselves. These results highlighted an increase in inhibition halo sizes when isolate 14 was co-cultivated with S. aureus or S. ser. Typhimurium. The recorded pH values between 8 and 20 h and 26 and 34 h of incubation time ranging from 6.6 to 6.8 for each batch culture. Comparative studies described the growth of S. aureus under much more stringent conditions, at pH values of 5.25 and 4.48 [58] or 5.6 in milk [59]. The inhibition of S. aureus in buffered media by Lactococcus strains was also observed at pH 6.8 and cannot, therefore, be attributed to a pH drop since S. aureus can grow at lower pH values [59]. Mufandaedza et al. also reported the survival and the growth of Salmonella sp. in fermented product associated with low pH [60].
The web-based tool BAGEL4 was used to mine secondary metabolite gene clusters within the whole genome L. lactis L3 (isolate 14). A coding sequence sharing 51.35% identity with the lcnB gene was recovered. The gene is known to be involved in the production of the lactococcin B bacteriocin (LcnB) [61]. A sequence sharing 69.23% identity with the lciB gene was also recovered and is involved in the immunity toward LcnB. A putative export system encoded by analogs of lcnC and lcnD genes was also identified with 98.37% identities. This analysis highlighted the presence of four genes sharing identities with the lactococcin operon [62]. The LcnB affects membrane permeability, dissipating its potential and its pH gradient caused by leakage of ions. In L. lactis strains carrying the immunity protein, the LcnB is inactive [63]. The LcnB antimicrobial activity was detected against close-related Lactococci strains. None of the Lactococci strains LcnB producers could inhibit the growth of other Gram-positive or Gram-negative bacteria [64,65]. Isolate 14 inhibited the growth of L. lactis IL1403, a plasmid-free strain sensitive to bacteriocin. These results suggested that isolate 14 could produce an antimicrobial peptide close to the LcnB.Şahİngİl et al. also reported inhibition of Salmonella sp. by a lactococcin-like bacteriocin, produced by an L. lactis subsp. lactis strain [66]. That lactococcin activity was observed during the early logarithmic phase (6-7 h of growth) and had maximal activity at 18 h of incubation time, as with the results shown here for isolate 14.

Conclusions
This study described the microbial community of an undefined starter culture with its richness in Lactococcus strains, naturally present in the raw milk microbiota. Five L. lactis variants were identified, and five carbohydrate metabolism profiles were highlighted in the Lactococcus genus. These Lactococcus strains differ in their phenotypic properties from the strains commonly used as industrial starters. Moreover, some L. lactis isolates presented an antimicrobial activity detected by cultural methods. These bacteria could act as a biocontrol agent when combined with other procedures in the so-called hurdle technology. Their GRAS status, their origin from an ecological niche and their role in flavor formation support the use of the UMSC in cheese manufacture procedures. Furthermore, considering the culture characteristics and their relationship with product quality, further studies could be helpful to determine their effects on technological aptitude and sensory properties of the cheese.

Supplementary Materials:
The following are available online at http://www.mdpi.com/2304-8158/9/5/622/s1, Figure S1. Carbohydrate metabolism tested using the API 50 CH system (BioMérieux, Marcy l'Etoile, France). A blue dot highlights a positive response to carbohydrate degradation. Figure S2. Pairwise average nucleotide identity (ANI) values across six Lactococcus genomes and the reference genome L. lactis subsp. lactis IL1403. Colors in the heatmap represent pairwise ANI values, with a gradient from blue (low identity) to green (high identity). Table S1. Antimicrobial spectrum of LAB candidate isolates against related LAB, food-spoilage and foodborne pathogens evaluated by the agar well diffusion assay. Plates were recorded for inhibition halos (mm).