Lactic Acid Bacteria Adjunct Cultures Exert a Mitigation Effect against Spoilage Microbiota in Fresh Cheese

Lactic acid bacteria (LAB) have a strong mitigation potential as adjunct cultures to inhibit undesirable bacteria in fermented foods. In fresh cheese with low salt concentration, spoilage and pathogenic bacteria can affect the shelf life with smear on the surface and packaging blowing. In this work, we studied the spoilage microbiota of an Italian fresh cheese to find tailor-made protective cultures for its shelf life improvement. On 14-tested LAB, three of them, namely Lacticaseibacillus rhamnosus LRH05, Latilactobacillus sakei LSK04, and Carnobacterium maltaromaticum CNB06 were the most effective in inhibiting Gram-negative bacteria. These cultures were assessed by the cultivation-dependent and DNA metabarcoding approach using in vitro experiments and industrial trials. Soft cheese with and without adjunct cultures were prepared and stored at 8 and 14 °C until the end of the shelf life in modified atmosphere packaging. Data demonstrated that the use of adjunct cultures reduce and/or modulate the growth of spoilage microbiota at both temperatures. Particularly, during industrial experiments, C. maltaromaticum CNB06 and Lcb. rhamnosus RH05 lowered psychrotrophic bacteria of almost 3 Log CFU/g in a 5-week stored cheese. On the contrary, Llb. sakei LSK04 was able to colonize the cheese but it was not a good candidate for its inhibition capacity. The combined approach applied in this work allowed to evaluate the protective potential of LAB strains against Gram-negative communities.


Introduction
Food spoilage is considered an important factor affecting the postharvest food losses, defined as the amount of edible food that is available for human consumption but is not consumed for any reason. The dairy products are the third food category in terms of the estimated total value of food losses and waste at retail and consumer levels. Gustavsson et al. [1] estimated that more than 20 and 12% of the initial milk and dairy production is lost or wasted respectively in the US and in Europe, and most of these losses occur at processing or consumption stages [2]. The main causes of cheese spoilage are microorganisms (bacteria, yeasts, and molds) that can grow during the manufacturing process or all along the shelf life, modifying sensorial properties, aroma, structure, and color of foodstuffs. While clostridia are the main cause of spoilage in long ripened hard cheeses [3,4], in fresh Lactobacilli were grown in anaerobic condition at 37 • C for 48 h in MRS broth at pH 6.5 (Becton, Dickinson and Company, Franklin Lakes, NJ, USA), Carnobacterium in anaerobic condition at 37 • C for 48 h in MRS broth at pH 8.5, and the Gram-negative bacteria in aerobic condition at 30 • C for 24 h in LB broth medium (Oxoid, Hampshire, UK).

Sampling and Microbiological Analyses
Fresh cheeses with 0.65% salt content stored at 8 • C for 25 days, one week after the usual shelf life, were collected from a dairy farm. Microbiological analyses were made on 15 batches produced in summer and 15 during autumn, collecting three cheeses from each production. Samples were processed as follows: 10 g of cheese were homogenized with 90 mL of 2% sodium citrate in a Stomacher Lab-Blender 400 (Laboratory Blender Seward, London, UK), tenfold diluted in peptone water, and plated on different selective media. Plate counts were done for the following microorganisms using different selective media: staphylococci on mannitol salt agar (Oxoid, Hampshire, UK) at 37 •

DNA Fingerprinting and Taxonomical Identification
DNA was extracted from selected colonies from all media using Microlysis (Clent Life Science, Amblecote Stourbridge, UK). RAPD PCR typing method was performed with RAPD2 primer (5 -AGCAGCGTGG-3 ) as described by Cocconcelli et al. [22] and REP PCR was carried out with GTG5 primer (5 -GTG GTG GTG GTG GTG-3 ) as described by Woods et al. [23]. PCR was performed in Mastercycler EPgradient S thermocycler (Eppendorf, Hamburg, Germany). PCR products were separated by electrophoresis at 100 V in a 2% agarose gel and stained with Syber Safe (Invitrogen, Carlsbad, California, US). The 16S rRNA gene was amplified using the primers P1 (5 -GCGGCGTGCCTAATACATGC-3 ) and P6 (5 -CTACGGCTACCTTGTTACGA-3 ). P1 was used to determine the partial 16S rRNA gene sequence. The 16S rRNA gene sequences were compared with the sequences present in the small sub-unit database (SSU-Prok) of Ribosomal Database Project Release 10 [24]; to obtain taxonomical and similarity rank calculations (S_ab) [25].

Inhibition Assays
A total of 14 adjunct cultures were tested for their inhibitory activity using MRS medium and fresh cheese agar (FCA) ( Table 1). FCA was composed by 120 g/L of fresh cheese and 20 g/L of agar. After homogenization, the medium was treated at 90 • C for 3 min to melt and pasteurized the medium. Adjunct cultures (10 7 CFU/mL) were inoculated in the agar media before pouring the plates, then, decimal dilution of the spoilage strains (10 µL) were spotted over the agar surface and plates incubated in aerobic condition at 8 • C and 14 • C for 48 h. As control, MRS and FCA plates without adjunct cultures but containing S. thermophilus ST022 (Clerici-Sacco Group, Como, Italy), as the solely starter culture, were used. The inhibitory activity was detected by assessing the count reduction of the spoilage strains on the media containing the adjunct cultures when compared with the control.

Industrial Trial
The studied cheese is a fresh "primosale" cheese, made with low fat pasteurized cow milk. In brief, the cheese-making process was the following: milk was pasteurized for 30 s at 78 • C, coagulation occurred at 37 • C after the addition of rennet, the curd was molded and then brined by immersion in a saturated salt solution (0.65% final salt concentration). Cheese was packaged in a modified atmosphere (70% N 2 and 30% CO 2 ) and stored at 4 • C.
The final pH of the cheese was 5.8. The industrial trial was based on three subsequent production cycles of fresh cheeses, used as replicates, and manufactured in the same cheese plant.
The experimental design was as follows: a control (STD) made using S. thermophilus ST022 as a starter culture was inoculated at a dose of 10 7 CFU/mL of milk. Treated cheeses, manufactured as the control, were produced by adding 5 × 10 6 CFU/mL of each different adjunct cultures, Llb. sakei LSK04 (SA), Lcb. rhamnosus RH05 (RH), C. maltaromaticum CNB06 (CB) (Clerici-Sacco Group, Como, Italy) in single or a mix of them (MX) where the three strains were present in equal amounts. Each treatment was manufactured in parallel in a different vat. In each replicate, half of the production was used for a challenge study, where cheeses were spiked, just before the packaging, by spraying onto the cheese surface a mixture of 10 spoilage isolates (Table 1) with a final concentration of 1 × 10 2 CFU/cm 2 . Spiked samples were named respectively STD1, SA1, RH1, CB1, MX1. Cheese samples packaged under a modified atmosphere composed by 70% N 2 and 30% CO 2 (MAP), were stored for 5 weeks at 8 and 14 • C, simulating, in the second case, a thermal abuse.

DNA Extraction
DNA was extracted from cheese samples according to the following protocol: 200 mg of cheese paste were collected in triplicate, processed with sodium citrate 2%, incubated for 10 min at 45 • C, and centrifuged at 8000 rpm for 8 min. Then, 40 µL of proteinase K and 400 µL of lysis buffer were added to the cell pellet and the tubes were then incubated at 56 • C with gentle shaking for 2 h. DNA extraction was performed using the Spin Tissue Mini Kit Invisorb (Invitek, Berlin, Germany) following the manufacturer's instructions. The elution step was repeated twice to increase the final DNA yield.

16S rRNA Amplicon Library Construction, Sequencing, and Bioinformatics
The V3-V4 regions of 16S rRNA gene were amplified using 331F primer: 5 -TCCTACGGGAGG CAGCAGT-3 and 797R: 5 -GGACTACCAGGGTATCTAATCCTGTT [26]. Primers were modified with forward overhang: 5 -TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG -[locus-specific sequence]-3 and with reverse overhang: 5 -GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAG-[locus-specific sequence]-3 , necessary for dual index library preparation. After indexing and pooling steps, libraries were loaded on Illumina MiSeq (Illumina Inc., San Diego, CA, USA) and sequenced with 2 × 300 bp paired end approach. Sequencing reads filtered for average quality (Q > 30) and R1 and R2 were merged using FLASH with default parameters [27]. Biological replicates were pooled. Qiime v 1.9.1. [28] was used to perform the full analysis from OTU picking to the statistical analysis, using the pick_closed_reference_otus wrapper for OTU picking with UCLUST against RDPII database trainset14_032015. OTUs were filtered at 0.005% abundance [29] and representative sequences were identified using SeqMatch ver.3 release 11, retaining the best KNN match. Alpha and beta diversity were studied by means of QIIME 1.9.1 using the alpha_diversity and beta_diversity_through_plots wrappers. Composition and diversity were analyzed using MicrobiomeAnalyst softwares [30]. PICRUSt (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States, http://picrust.github.io/picrust) was used to predict the functional profiles of the microbial communities. QIIME 1.8 was used to pick the OTUs at 97% identity against the Greengenes database (version 05/2013) with the closed reference method. Statistical Analysis of Metagenomic Profile software package (STAMP) was used for the statistical analysis. Significant differences among all samples were calculated using ANOVA test along with Tukey-Kramer test with p values (p < 0.05) [31].

Nucleotide Sequence Accession Numbers
Reads were deposited in the Sequence Read Archive (SRA) database under accession number SRP071740.

Characterization of Fresh Cheese Spoilage Microbiota by Culture-Dependent Approach
The microbiological counts performed on samples collected one week after the end of the shelf life showed significantly higher counts in summer batches than in cheese manufactured during the autumn, for all the considered bacterial groups, with the only exception of S. thermophilus, which was present in comparable numbers, being used as starter culture in both periods, as shown in Figure 1. trainset14_032015. OTUs were filtered at 0.005% abundance [29] and representative sequences were identified using SeqMatch ver.3 release 11, retaining the best KNN match. Alpha and beta diversity were studied by means of QIIME 1.9.1 using the alpha_diversity and beta_diversity_through_plots wrappers. Composition and diversity were analyzed using MicrobiomeAnalyst softwares [30]. PICRUSt (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States, http://picrust.github.io/picrust) was used to predict the functional profiles of the microbial communities. QIIME 1.8 was used to pick the OTUs at 97% identity against the Greengenes database (version 05/2013) with the closed reference method. Statistical Analysis of Metagenomic Profile software package (STAMP) was used for the statistical analysis. Significant differences among all samples were calculated using ANOVA test along with Tukey-Kramer test with p values (p < 0.05) [31].

Nucleotide Sequence Accession Numbers
Reads were deposited in the Sequence Read Archive (SRA) database under accession number SRP071740.

Characterization of Fresh Cheese Spoilage Microbiota by Culture-Dependent Approach
The microbiological counts performed on samples collected one week after the end of the shelf life showed significantly higher counts in summer batches than in cheese manufactured during the autumn, for all the considered bacterial groups, with the only exception of S. thermophilus, which was present in comparable numbers, being used as starter culture in both periods, as shown in Figure 1. This observation was consistent with the notice that summer samples, after a prolonged shelf life (one week after expiration date), presented a higher incidence of surface smear and gas defects (data not shown). In particular, the spoilage microbiota of summer production was dominated by bacteria growing on selective media for Pseudomonas (6.9 Log CFU/g), heterofermentative This observation was consistent with the notice that summer samples, after a prolonged shelf life (one week after expiration date), presented a higher incidence of surface smear and gas defects (data not shown). In particular, the spoilage microbiota of summer production was dominated by bacteria growing on selective media for Pseudomonas (6.9 Log CFU/g), heterofermentative Lactobacillus (6.7 Log CFU/g), and Enterobacteriaceae (5.8 Log CFU/g), while other bacterial groups were found in lower numbers. To better identify the spoilage microbiota, 200 randomly selected colonies from agar plates of the different media and from both seasons were analyzed by RAPD-PCR and REP-PCR to determine the prevalence of single spoilage strains in the cheese products and the possible persistence in the production environment. RAPD and REP-PCR fingerprinting analysis recognized 48 different strains that were subsequently identified by 16S rRNA gene sequence analysis (Table 2). The complex microbial community of sampled cheeses was dominated by Obesumbacterium proteus, which was detected on both VRBA and Pseudomonas agar plates. The most prevalent strains of this species were isolated from both summer and autumn manufactures ( Table 2), suggesting their persistence in the cheese plant. In lower numbers, also strains belonging to Aerococcus urinaeequi and Enterococcus faecalis have been found in the cheese samples from summer production. In addition, different species of Pseudomonas genus were detected: Pseudomonas libanensis, P. fragi, and P. psychrophila uniquely identified in the summer samples while P. brennerii, P. gessardii, and P. proteolytica differently found in autumn's cheeses.

In Vitro Selection of Adjunct Cultures
The second purpose of this study was to ascertain if adjunct cultures of LAB were able to reduce the detected spoilage defect in fresh cheese. To identify the most suitable cultures, 14 different strains (Table 1), known for their ability to produce antimicrobial compounds, were evaluated (data not shown). These strains were tested on agar plates and in cheese agar models against ten strains isolated from the spoiled fresh cheeses object of this study (Table 1). Results showed that the most active strains at the two temperatures of incubation (8 and 14 • C) were Llb. sakei LSK04, Lcb. rhamnosus LRH05, and C. maltaromaticum CNB06 on both used media (MRS and FCA). These cultures induced a reduction of more than 3 Log CFU when compared to the control strain. Additionally, the three selected strains were tested against each other to assess the possibility of cross inhibition when used together in mixed culture. No evident effect of cross inhibition was detected, being the reduction of viable cells below 1 Log CFU. Consequently, these three strains were selected for further studies in industrial conditions.

Efficacy Assessment of Adjunct Cultures in Industrial Conditions
To study the effect of the storage temperature on the spoilage microbiota development during shelf life and to assess the spoilage mitigation in fresh cheese, a trial in an industrial plant was performed, reproducing for the shelf life period at 8 • C, the average temperature of domestic refrigerators [32], and a thermal abuse at 14 • C. The microbiological plate counts on selective media for Enterobacteriaceae and Pseudomonas from three independent productions for each condition are reported in Figure 2.
viable cells below 1 Log CFU. Consequently, these three strains were selected for further studies in industrial conditions.

Efficacy Assessment of Adjunct Cultures in Industrial Conditions
To study the effect of the storage temperature on the spoilage microbiota development during shelf life and to assess the spoilage mitigation in fresh cheese, a trial in an industrial plant was performed, reproducing for the shelf life period at 8 °C, the average temperature of domestic refrigerators [32], and a thermal abuse at 14 °C. The microbiological plate counts on selective media for Enterobacteriaceae and Pseudomonas from three independent productions for each condition are reported in Figure 2.  For what concerns the STD samples, no evident effect of temperature was observed in the plate counts on the two selective media, despite the lower pH values in control samples at 14 • C than those stored at 8 • C (5.3 and 5.9 respectively). This pH reduction was associated with an increase in lactic acid bacteria counts when samples were stored at 14 • C. When cheese samples inoculated with the adjunct cultures were analyzed applying the cultivation-dependent approach, the effect on the spoilage microbiota was observed at both temperatures and in presence or absence of spiking (Figure 2). At 8 • C, a significant reduction of spoiling bacteria when compared to STD control samples, made with the starter S. thermophilus alone, was observed. Counts on VRBA decreased of 2.0 Log CFU/g and Pseudomonas agar of 2.3 Log CFU/g CB cheeses produced with C. maltaromaticum CNB06 alone and more than 1 Log CFU/g on both media was obtained in MX cheeses produced with a mixed combination of the three protective strains, (Figure 2A). In cheeses intentionally spiked with the mix of isolated spoilage strains, this effect was higher, being the mitigation effect evident not only in CB1 and MX1 samples but also in RH1 samples inoculated with Lcb. rhamnosus RH05 respect to STD1 cheese made without adjunct cultures (Figure 2A). Similarly, in cheeses stored at 14 • C, Lcb. rhamnosus RH05, C. maltaromaticum CNB06, and the mix of the three adjunct cultures significantly decreased the counts of spoilage bacteria in both natural and spiked treatments ( Figure 2B). Samples CB, MX, CB1, RH1, and MX1 at 8 • C and CB, RH, MX, RH1, and MX1 at 14 • C did not present evident sign of spoilage, discoloration, and package blowing. Metagenomic results in terms of relative abundance of taxonomic units at species level for cheese samples with protective cultures stored at 8 and 14 • C is shown in Figure 3.
High-throughput 16S rRNA gene sequencing, applied to assess mainly the changes induced by LAB cultures on the microbial community of cheeses in the two experimental conditions, resulted in 14,756,157 raw reads that were automatically filtered to 13,565,705 by the Miseq software. After the Flash step, the extended frags were 9,346,660 with an average length of 465 bp (± 6.6) bp. Biological replicates were pooled to a total of 7,578,763 reads and 542 OTUs were obtained after OTU picking and filtering steps. The 16S rRNA gene metabarcoding done on one independent production gave a total of 465,040 reads for STD at 8 • C and 413,106 for STD at 14 • C with an average length of 465 bp (± 6.6). The analysis revealed 382 different OTUs in STD at 8 • C and 411 in STD at 14 • C. In terms of phyla, the bacterial community of STD at 8 • C was constituted by approximately 62% of Proteobacteria and 35% of Firmicutes, while in STD at 14 • C Proteobacteria decreased to 40% and Firmicutes raised to 59%. The dominant species detected in STD at 8 • C are reported in Figure 3A: Gram-positive species were the starter culture S. thermophilus and Lactococcus lactis, which accounted respectively to 8% and 22% of detected amplicons. Spoilage bacteria represented the dominant group, being Acinetobacter johnsonii the 25% and members of the Enterobacter genus the 20% of the total microbial community. At 14 • C ( Figure 3B), the proportion of single species changed: Lactococcus lactis decreased (7%), S. thermophilus increased (38%), and L. plantarum, Llb. Sakei, and Lcb. rhamnosus were detected. The spoilage microbiota did not significantly change its composition. In none of the analyzed samples, members of the genus Pseudomonas were detected, demonstrating the effect of MAP in controlling this bacterial group. As a matter of fact, positive counts found on Pseudomonas agar base were mostly characterized as members of the Enterobacteriaceae family.
According to the meta-barcoding approach, the spread on the cheese surface of the ten spiking strains did not affect the natural species composition of the analyzed samples; only the spiking species Obesumbacterium proteus was detected in low amounts in SA1 samples. Consistently, the alpha-diversity index of spiked and non-spiked samples was not statistically different, as shown in Figure 4A. No difference in the alpha-diversity index of samples incubated at 8 and 14 • C was also observed ( Figure 4B), while the use of adjunct cultures had statistically significant effect (p < 0.05) on cheese microbiota at both temperatures of storage ( Figure 4C).
The three species added as adjunct cultures were able to colonize the cheese, being detected as dominant populations either when used as single strain or mixed culture (Figure 3). At 8 • C, the most effective strain was C. maltaromaticum CNB06, which reduced the number of amplicons related to Gram-negative bacteria below 2% as well as the strains mix, particularly in the spiked samples, confirming the plate counts data (Figure 2). At 14 • C, all the tested adjunct cultures reduced the prevalence of Gram-negative species and in treatments with adjunct cultures MX, CB1, and MX1 at 14 • C also an increase of C. tyrobutyricum was observed.   strains did not affect the natural species composition of the analyzed samples; only the spiking species Obesumbacterium proteus was detected in low amounts in SA1 samples. Consistently, the alpha-diversity index of spiked and non-spiked samples was not statistically different, as shown in Figure 4A. No difference in the alpha-diversity index of samples incubated at 8 and 14 °C was also observed ( Figure 4B), while the use of adjunct cultures had statistically significant effect (p < 0.05) on cheese microbiota at both temperatures of storage ( Figure 4C). The three species added as adjunct cultures were able to colonize the cheese, being detected as dominant populations either when used as single strain or mixed culture (Figure 3). At 8 °C ,the most effective strain was C. maltaromaticum CNB06, which reduced the number of amplicons related to Gram-negative bacteria below 2% as well as the strains mix, particularly in the spiked samples, confirming the plate counts data (Figure 2). At 14 °C , all the tested adjunct cultures reduced the prevalence of Gram-negative species and in treatments with adjunct cultures MX, CB1, and MX1 at 14 °C also an increase of C. tyrobutyricum was observed.

Effect of Adjunct Cultures on the Functional Profiling of Cheese Microbiome
The functional profiling of cheese microbiome, based on functionality prediction, revealed significant differences in KEGG orthologues when the control cheese samples, stored at 8 and 14 °C, were compared to samples with adjunct cultures. Among the affiliated KEGG pathways that reached a statistically significance (p < 0.01), 21 pathways were selected, with relevance for the bacterial metabolism in cheese as reported in Figure 5.

Effect of Adjunct Cultures on the Functional Profiling of Cheese Microbiome
The functional profiling of cheese microbiome, based on functionality prediction, revealed significant differences in KEGG orthologues when the control cheese samples, stored at 8 and 14 • C, were compared to samples with adjunct cultures. Among the affiliated KEGG pathways that reached a statistically significance (p < 0.01), 21 pathways were selected, with relevance for the bacterial metabolism in cheese as reported in Figure 5. In cheese stored at 14 °C, genes coding for carbohydrate metabolism, including sugar phosphotransferase system (PTS), galactose metabolism, glycolysis, and pentose phosphate pathway increased if compared to samples at 8 °C. Differently, at the highest temperature, a significant reduction of genes involved in the amino acids metabolism and degradation was observed.
The effect on bacterial metabolism produced by the different protective strains in terms of functional categories is reported in Figure 6. In cheese made with C. maltaromaticum ( Figure 6C), 17 out of 21 affiliated KEGG pathways were modulated; in particular, sugar metabolism pathways  In cheese stored at 14 • C, genes coding for carbohydrate metabolism, including sugar phosphotransferase system (PTS), galactose metabolism, glycolysis, and pentose phosphate pathway increased if compared to samples at 8 • C. Differently, at the highest temperature, a significant reduction of genes involved in the amino acids metabolism and degradation was observed.
The effect on bacterial metabolism produced by the different protective strains in terms of functional categories is reported in Figure 6. In cheese made with C. maltaromaticum ( Figure 6C), 17 out of 21 affiliated KEGG pathways were modulated; in particular, sugar metabolism pathways (galactose metabolism, PTS system, glycolysis, pentose phosphate pathway, and pyruvate metabolism) increased, while amino-acids metabolism (tryptophan; lysine; arginine, and proline; valine, leucine, and isoleucine; glycine, serine, and threonine), together with lipopolysaccharides biosynthesis and nitrogen metabolism decreased. Similarly, Lcb. rhamnosus RH05 ( Figure 6B) had an effect on the cheese microbiome, increasing the energetic metabolism pathways, including galactose metabolism, pyruvate metabolism, PTS system, pentose phosphate pathway, and glycolysis. Although Llb. sakei LSK04 ( Figure 6A) and the mix ( Figure 6D) showed a more limited effect than the other two single protective strains, the trend was similar and a reduction in amino acids metabolism, including tryptophan, lysine, arginine and proline, glycine, serine, threonine, and branched chain amino acids, was observed.

Discussion
The increasing demand of low-salt and low-fat cheese requires new hurdle technologies approaches to reduce the spoilage microbial communities along the shelf life, particularly by psychotropic Gram-negative bacteria [33,34] that can also trigger antimicrobial resistances [35]. To address this issue, this study focused on the characterization of the spoilage microbiota of these dairy products and the appraisal of lactic acid bacteria adjunct cultures in limiting the outgrowth of these undesirable microorganisms and consequently cheese defects.
The composition of the microbial community of a low salt fresh cheese (0.65% salt, pH 5.8) was examined after an extended shelf life, when defects such as gas production, surface discoloration, and off-flavors are usually present. This product was examined twice: first, samples collected in two different seasons, and then the samples used as control during the industrial trial. In the 30 cheese samples, selected after the end of the shelf life from both summer and autumn periods of production, psychotropic Gram-negative bacteria, Enterobacteriaceae and Pseudomonas, represented the dominant population of spoiled cheeses, as already observed [33,34].
When bacterial strains were isolated and identified, Obesumbacterium proteus was the most frequent, accounting for more than 40% of isolates in both seasons; this microorganism is usually recognized as a brewery contaminant [36] but was also found in cheese [37]. Other bacterial spoilers found at a lower frequency belonged to Aerococcus, Enterococcus, and Pseudomonas. Hafnia alvei, a well-known component of smear-ripened cheese [38] that can affect aromatic features by producing sulfur compounds and free amino acids, was also found.
When the microbiota of fresh cheese was investigated by DNA metabarcoding approach, the Gram-negative spoilage microbiota represented the 90% of the total microbial community, with the dominance of Acinetobacter and Enterobacter. Differently from what is observed in other studies [39,40], Pseudomonas were not found in the control samples, neither in the other samples using DNA metabarcoding analysis, confirming the effectiveness of the MAP. Although defects were still present, control cheeses stored at 14 • C presented a higher diversity in terms of OTU, with a decreased proportion of Enterobacteriaceae and an increased prevalence of Firmicutes (40%), mainly S. thermophilus and Lactobacillus species. This can be explained by a more permissive storage temperature that can engage competitive interactions between the spoilage microbiota and LAB. The effect of temperature is also highlighted by the predicted functional profiling of cheese microbiome where samples stored at 14 • C presented a significant reduction in the metabolic pathways, such as those involved in amino acid degradation, that can lead to unpleasant organoleptic defects.  With the aim of reducing the identified spoilage microbiota in industrial manufacturing, it was necessary to test the bio-preservative effect of different protective cultures in a fresh cheese through a combination of culture-based and next generation sequencing approaches. Differently from previous studies, where the effect of commercial or selected protective cultures was tested directly during cheese manufacturing [41,42], adjunct LAB cultures were firstly selected, which could specifically inhibit the spoilage microbiota from the cheese under examination and subsequently trialed in industrial cheese manufacturing. The preliminary screening in plate and in cheese models allowed to select three LAB strains, Llb. sakei LSK04, Lcb. rhamnosus RH05, and C. maltaromaticum CNB06 that reduced of almost 3 Log CFU/g Enterobacteriaceae and Pseudomonadaceae populations at 8 and 14 • C in a cheese model. The scale-up at industrial level was made adding the three selected strains, alone or in pool, directly in milk for cheese production and studying the development of spoilage during the storage at 8 and 14 • C under MAP, a technology widely used to improve the shelf life of fresh dairy products, limiting the growth of pathogenic and alterative microorganisms [34,43].

Discussion
At 8 • C, the temperature most frequently detected in consumers' refrigerators (EFSA, 2008), all protective strains were able to grow in the cheese matrix, demonstrating a good adaptation to the substrate. C. maltaromaticum CNB06 and, consequently the mixed culture, showed the highest protective action, reducing of more than 2 Log CFU/g the rate of undesired microorganisms respect to the control cheese. In addition, in samples where a mix of Enterobacteriaceae and Pseudomonas was intentionally spiked by spraying on cheese surfaces, C. maltaromaticum CNB06 confirmed its inhibitory effect, lowering the spoiling bacteria between 2-3 Log CFU/g. Food-isolated C. maltaromaticum has been extensively studied for its inhibitory capacity due to bacteriocin production, particularly against Listeria monocytogenes [44,45]; moreover, this bacterium plays a positive role in soft cheese making thanks to the ability to be acid tolerant without affecting the starter activity during coagulation, and conferring also a pleasant malty aroma to the final product [33,42,46]. Additionally, in spiked samples, Lcb. rhamnosus RH05 showed a significant reduction of the spoilage microbiota. In case of thermal abuse (14 • C), the protective effect of selected strains was even higher due to an improved growth of LAB at this temperature with consequent pH lowering to average values below 5.0. In this condition, the efficacy of the protective treatments, with a 2 Log CFU/g reduction, was demonstrated for C. maltaromaticum CNB06, Lcb. rhamnosus RH05, and the combination of the three strains.
When the DNA metabarcoding data were analyzed for the cheese samples with the addition of protective cultures, the results obtained by the cultivation-dependent approach were confirmed. The inhibitory potential at 8 • C of C. maltaromaticum CNB06 was demonstrated; in both CB and CB1 cheese samples, the Gram-negative population was below the 5% of the total OTUs, being the microbiota mainly composed by C. maltaromaticum, S. thermophilus, Llb. sakei, and L. lactis. This proved that C. maltaromaticum CNB06 was well adapted to fresh cheese and was able to colonize the dairy environment at refrigeration temperature even when directly added to milk in the vat for cheese making. Spanu et al. [42] previously confirmed the protective potential of this psychrophilic microorganism to control the growth of spiked Pseudomonas spp. when inoculated on the surface of ricotta fresca cheese and supported previous studies conducted on naturally contaminated sheep's milk ricotta fresca [33]. Moreover, C. maltaromaticum demonstrated to increase the sugar metabolism in cheese at the end of shelf life despite the decrease of amino acids and lipopolysaccharides metabolisms; this is in line with the high numbers of carbohydrate transport and metabolism genes found on the chromosome of Carnobacterium spp. previously sequenced [47]. In addition, the inhibitory capacity of C. maltaromaticum seems to be correlated to bacteriocins synthesis but also to the production of formate and acetate organic acids [48]. Therefore, its growth is not associated to high pH variations at the end of the shelf life respect to the standard sample (5.56 versus 5.76).
When Lcb. rhamnosus RH05 was added as protective culture, more than 50% of the taxonomic units were assigned to this LAB species at 8 • C and 30% at 14 • C. In spiked RH1 samples, the spoilage microbiota accounted for less than 30%, being represented by Acinetobacter and Enterobacter. The protective effect of this strain, although less effective than C. maltaromaticum CNB06, allowed to reduce unwanted microorganisms and delay the product spoilage. Lcb. rhamnosus RH05 in a blend with L. plantarum and E. faecium strains had a low mitigation activity against Pseudomonas in ricotta fresca cheese [33]. Differently, when the same strain was tested with the solely L. plantarum in burrata cheese [49], it slowed the growth of staphylococci, coliforms, and Pseudomonas spp., especially in early storage, increased the product shelf life of three days, and improved the flavor. Furthermore, the role of Lcb. rhamnosus as a promising protective culture was widely demonstrated in dairy products [50,51] and also in food coating applications where cell-free supernatant of Lcb. rhamnosus were effective in inhibiting E. coli, L. monocytogenes, S. aureus, or Salmonella Typhimurium when used with whey protein films in packaging [52].
Although the cultivation-based data did not show a significant reduction of the spoilage counts, metabarcoding outcomes in samples stored at both temperatures, when Llb. sakei LSK04 was used as the only adjunct culture, indicate that this bacterium colonized the cheese during the shelf life. According to our experiments in fresh cheese, Llb. sakei LSK04 was not a good candidate for its inhibition capacity among the tested microorganisms, since, also in condition of thermal abuse, the strain was not able to provide a good level of protection, even if prevailed in SA and SA1 samples as taxonomic unit on the total OTU abundance.
Latilactobacillus sakei is more frequently used as a protective agent in meat and meat products to control Listeria monocytogenes and other pathogenic bacteria [53,54] thanks to its ability to grow in cold storage conditions and driving, at the same time, the fermentation process. When this species is forced to grow in a different food environment, such as cheese, it can sometimes vary its own efficacy [55].
When the three adjunct cultures were evaluated as a pool, positive results were obtained at 8 and 14 • C mostly due to the presence of Carnobacterium, but also to the synergic effect generated by bacteriocins production by the three bacterial strains. LAB mixed cultures often demonstrated to exert a highest protective action without affecting the cheese making and the final organoleptic properties [14,15,56]. As a general observation, the use of adjunct cultures significantly shapes the cheese microbiota, by reducing the alpha-diversity index, as shown in Figure 4. This is the consequence of the inhibitory activity played by the studied LAB against Gram-negative bacteria.
In this study, we have analyzed the same samples using different approaches, plate counts, and OTU detection by metabarcoding. Although these two techniques are known to provide a different estimation of the bacterial community composition [57], the results obtained in this study are consistent, showing that the decrease of 2-log CFU on selective media corresponds to approximately a 90% reduction of the corresponding spoiling population in OTU analysis.
The use of bacterial cultures as protection against spoiling and pathogenic microorganisms represents a natural and affordable intervention in the food industry. Our work provides additional data supporting the efficacy of selected adjunct cultures in limiting the Gram-negative spoilage populations in fresh cheese by reshaping the cheese microbiome.

Acknowledgments:
The autors would like to acknowledge any support given which is not covered by the author contribution or funding sections. This may include administrative and technical support, or donations in kind (e.g., materials used for experiments).

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.