Ecology of Lactobacilli Present in Italian Cheeses Produced from Raw Milk

The microbiota is known to play a key role in the development of the organoleptic features of dairy products. Lactobacilli have been reported to represent one of the main components of the nonstarter bacterial population, i.e., bacteria that are not deliberately added to the milk, harbored by cheese, although the species-level composition of this microbial population has never been assessed in detail. In the present study, we applied a recently developed metagenomic approach that employs an internally transcribed spacer to profile the Lactobacillus population harbored by cheese produced from raw milk at the (sub)species level. The obtained data revealed the existence of particular Lactobacillus community state types consisting of clusters of Lactobacillus (sub)species that tend to cooccur in the screened cheeses. Moreover, analysis of covariances between members of this genus indicate that these taxa form an elaborate network of positive and negative interactions that define specific clusters of covariant lactobacilli.

ABSTRACT

Among the bacterial genera that are used for cheese production, Lactobacillus is a key taxon of high industrial relevance that is commonly present in commercial starter cultures for dairy fermentations. Certain lactobacilli play a defining role in the development of the organoleptic features during the ripening stages of particular cheeses. We performed an in-depth 16S rRNA gene-based microbiota analysis coupled with internally transcribed spacer-mediated Lactobacillus compositional profiling of 21 common Italian cheeses produced from raw milk in order to evaluate the ecological distribution of lactobacilli associated with this food matrix. Statistical analysis of the collected data revealed the existence of putative Lactobacillus community state types (LCSTs), which consist of clusters of Lactobacillus (sub)species. Each LCST is dominated by one or two taxa that appear to represent keystone elements of an elaborate network of positive and negative interactions with minor components of the cheese microbiota. The results obtained in this study reveal the existence of peculiar cheese microbiota assemblies that represent intriguing targets for further functional studies aimed at dissecting the species-specific role of bacteria in cheese manufacturing.

IMPORTANCE The microbiota is known to play a key role in the development of the organoleptic features of dairy products. Lactobacilli have been reported to represent one of the main components of the nonstarter bacterial population, i.e., bacteria that are not deliberately added to the milk, harbored by cheese, although the species-level composition of this microbial population has never been assessed in detail. In the present study, we applied a recently developed metagenomic approach that employs an internally transcribed spacer to profile the Lactobacillus population harbored by cheese produced from raw milk at the (sub)species level. The obtained data revealed the existence of particular Lactobacillus community state types consisting of clusters of Lactobacillus (sub)species that tend to cooccur in the screened cheeses. Moreover, analysis of covariances between members of this genus indicate that these taxa form an elaborate network of positive and negative interactions that define specific clusters of covariant lactobacilli.

INTRODUCTION

Organoleptic properties of cheese depend on a complex network of chemical and enzymatic interactions between the cheese and its resident bacterial community (1), i.e., the cheese microbiota. A wide range of factors define cheese-associated organoleptic features, such as casein type, fat, and carbohydrates present in the milk used for cheese manufacturing, along with their biochemical modification performed by bacteria that colonize the cheese matrix, originating from environmental contamination or the use of unpasteurized milk or deliberately added as microbial starters (2). In this regard, the enzymatic degradation of caseins is directly or indirectly responsible for the release of flavor-defining components such as volatile aroma compounds (2, 3). Moreover, metabolism of lipids and lactose extensively contribute to texture, mouthfeel, taste, and specific flavors, especially in ripened cheeses, while additional biochemical activities may influence industrially relevant features such as nutrient composition and shelf life (1, 4).

Application of recently developed metagenomic techniques, facilitated by next-generation DNA sequencing approaches, has generated an in-depth view of the taxonomic composition of the cheese microbiota that is independent of cultivation and allows the detection of subdominant microorganisms (5–8). These studies have revealed that the cheese microbiota is a highly diverse population with sometimes unusual microbial dynamics linked to cheese-making practices and ripening (9–15). These studies have also highlighted that lactobacilli represent one of the main constituents of the cheese microbiota, as denoted by their high prevalence and average relative abundance in profiled cheeses (9–15).

Nonetheless, although the use of 16S rRNA gene-based microbial profiling determines the microbial composition of a given sample at genus-level accuracy, the subgenus ecology of the cheese microbiota has not yet been addressed in any significant detail. Indeed, 16S rRNA-based profiling does not allow assessment of diversity that may exist at the species level, thereby preventing accurate exploration of the microbiota composition of a given dairy product. For this reason, we applied a recently developed metagenomic methodology, i.e., Lactobacillus-based internally transcribed spacer (ITS) profiling (16). This approach relies on the ITS sequence, a ubiquitous genomic region with a much higher sequence diversity than the commonly used phylogenomic 16S rRNA marker gene. This ITS-based method allows accurate (sub)species profiling of members of the Lactobacillus genus present in samples corresponding to a range of traditional Italian cheeses that are produced from raw milk, i.e., milk that has not been pasteurized, and four ricotta cheeses, which are produced from whey.

RESULTS AND DISCUSSION

Dissecting the distribution of lactobacilli associated with Italian cheeses.

Twenty-one commercially sold Italian cheeses were sampled from the core section of the obtained cheese block and analyzed using a metagenomic approach aimed at studying the ecology of members of the Lactobacillus genus harbored by this food matrix (Table 1). We selected 17 cheeses manufactured from raw milk, i.e., not processed with pasteurization or other thermal treatments. Of these 17 cheeses, seven were made without the deliberate addition of any bacterial starter culture, and fermentation and ripening of these dairy products are therefore dependent on viable bacteria that are present in the raw milk and the fermentation environment (Table 1). In contrast, 10 cheeses are produced by supplementation with natural whey cultures (NWCs), i.e., undefined cheese-starter cultures obtained by the so-called back-slopping procedure (Table 1). Back-slopping is a process where whey from a previous cheese fermentation event is kept (for up to several days) and subsequently added to raw milk in order to initiate milk acidification and corresponding cheese production (17). In this context, it is worth mentioning that the bacterial population associated with NWCs tend to be dominated by one or just a small number of species, such as Lactobacillus helveticus, Streptococcus thermophilus, and Lactobacillus delbrueckii subsp. lactis, probably due to genetic adaptation to the milk/whey environment and to their thermophilic features that ensure the optimal growth of these microbial species at the temperature required for coagulation of the curd (ranging from 32 to 45°C) (17). In addition, we sampled four ricotta cheeses that are produced following a heat treatment of whey at high temperature (>80°C) for about 5 min, without the supplementation of any starter culture. Additional details of the sampled cheeses, such as production site, aging, and milk type, are reported in Table 1. Unfortunately, we were unable to retrieve information on the precise temperatures used for milk fermentation/coagulation of the curd, and each of these cheese producers may therefore have used a specific temperature (typically ranging from 32 to 45°C) irrespective of the type of cheese produced. Genus-level taxonomic profiles of these cheese samples were obtained by 16S rRNA gene-based microbial profiling, achieved by corresponding amplicon sequencing that generated a total of 973,861 reads, with an average of 46,374 reads per sample (Table 1).

TABLE 1.

Details about the cheeses used in this study and the quality data achieved for each sample by 16S rRNA profiling and Lactobacillus ITS rRNA profiling analyses

The collected and processed data sets were used to gain insight into the biodiversity of each sample by a bar plot representation constructed by employing the observed operational taxonomic unit (OTU) index (Fig. 1). Notably, all rarefaction curves obtained by means of subsampling (see Materials and Methods) were shown to reach a plateau, indicating that most biodiversity was captured by the applied analysis (see Fig. S1a in the supplemental material). Furthermore, inspection of the microbial biodiversity at 20,000 reads showed that cheeses 7, 9, and 10, all corresponding to fresh ricotta, were shown to exhibit the highest level of bacterial diversity (65, 61, and 86 OTUs, respectively), indicating that the particular procedure used to manufacture fresh ricotta, which involves the coagulation of whey proteins, supports an increased level of bacterial species diversity compared to other cheese types (Fig. S1a). It is possible that the high-temperature treatment required for the manufacturing of ricotta cheese reduced the fitness of otherwise cheese-dominant bacterial genera, in particular various lactic acid bacteria (LAB). The heat treatment therefore may have limited the ability of such LAB to acidify the environment and to release bacteriostatic compounds, which would consequently favor the proliferation of other accessory bacterial genera.

FIG 1.

Alpha-diversity analysis of 16S rRNA and Lactobacillus ITS microbial profiling. (a) Alpha-diversity based on 16S rRNA gene-based microbial profiling. (b) Lactobacillus ITS profiling-based alpha-diversity. The vertical axis reports the number of observed OTUs, while the horizontal axis displays the cheese samples. Each bar is colored in order to subsample it with the milk type used in the process. More specifically, the different colors of the bars reflect different milk types: light blue refers to buffalo milk, orange indicates milk from sheep, green represents goat milk, and red corresponds to cow milk.

Subsequently, beta-diversity analysis at the genus level, based on the Bray-Curtis index, was calculated and represented by means of a principal coordinate analysis (PCoA) (Fig. 2a), revealing the presence of two clusters, named cluster 16S-1 and cluster 16S-2, which encompass 12 and 5 samples, respectively, while the positions of the remaining cheese samples are scattered across this PCoA plot (Fig. 2a).

FIG 2.

Evaluation of beta-diversity in cheese samples. (a) Predicted PCoA based on 16S rRNA gene-based sequencing data using the Bray-Curtis index. (b) PCoA representation of a beta-diversity analysis performed for Lactobacillus ITS sequencing data using the Bray-Curtis index.

Based on these observations, we performed an in-depth analysis of all cheese samples that were included in each cluster in order to inspect their taxonomic composition at genus level for cluster-specific features, as based on 16S rRNA gene-mediated microbial profiling (Fig. 2; see also Tables S1 and S2 in the supplemental material).

Intriguingly, we observed that the genera Streptococcus and Lactobacillus represent >90% of the total average relative abundance of samples forming cluster 16S-1, with the single exception of sample 8, corresponding to aged ricotta, which was shown to contain a microbiota consisting of Staphylococcus at 38.2% relative abundance in addition to the dominant genera Streptococcus and Lactobacillus (Table S2; Fig. 3). Notably, these latter two genera have previously been reported to be dominant members of particular Italian cheeses, most likely due to their higher ecological fitness compared to other (lactic acid) bacterial competitors during the curd coagulation step, which occurs at temperatures of up to 45°C (17). In addition to the information previously reported, 5 of the 12 samples contained in this cluster were produced from buffalo milk, indicating a possible link between the cheese microbiota profile and the type of milk used for its production. Instead, cluster 16S-2 included samples in which the Lactococcus genus dominated the microbiota over other microbial taxa (>70%). Since Lactococcus has an optimal growth of 30°C, these findings indicate that cheeses constituting cluster 16S-2 were probably produced at lower temperatures compared to those constituting cluster 16S-1. Remarkably, four of the five cheeses constituting this group, with the exclusion of cheese 16 (fresh caprino), had been produced without the addition of any bacterial starter (including back-slopping), suggesting that the Lactococcus population harbored by raw milk has a major impact in defining the final cheese microbiota. Moreover, cheese samples falling outside these clusters are widespread across the whole PCoA representation (Fig. 2) and showed variable taxonomic profiles (Table S2; Fig. 3).

FIG 3.

Genus-level taxonomic composition of 16S rRNA sequencing-based PCoA cluster. (a) Genus composition of the samples constituting each cluster, with the relative abundance on the vertical axis and the sample on horizontal axis. (b) Average relative abundance of the three PCoA clusters. The average relative abundances of each bacterial genus and clusters are reported on the vertical and horizontal axis, respectively.

Intriguingly, permutational multivariate analysis of variance (PERMANOVA) between clusters 16S-1 and 16S-2 resulted in a P value of <0.01 (Table S3). Based on these data, we propose the existence of two main cheese community state types (CSTs), named cheese CST 1 and cheese CST 2, respectively, based on cheeses produced from raw milk. Notably, the distinctive feature that characterizes cheese CST 1 and cheese CST 2 is the dominance of cheese microbiota by Streptococcus/Lactobacillus or Lactococcus, respectively. Intriguingly, Streptococcus/Lactobacillus are thermophilic LAB with optimal growth temperatures at 40 to 42°C, whereas Lactococcus is a mesophilic bacterium with an optimal growth temperature of 30°C, indicating a correlation between CSTs and the particular “cooking” temperature used by cheese makers to manufacture a particular cheese product. Further studies that will involve a broader assortment of cheeses with known technological details regarding the production process will be of high scientific interest to confirm and extend the obtained results presented here. Furthermore, identification of cheeses falling outside these clusters (Table S2; Fig. 2a and 3) indicates the existence of an additional cheese CST with a lower prevalence among raw milk-based cheeses. In this context, a common feature observed for these four unclustered samples, three of which being produced without addition of bacterial starters (Table 1), is their high biodiversity (>40 observed OTUs).

Due to the industrial, scientific, and clinical/probiotic relevance of the genus Lactobacillus, along with the availability of a validated tool for reconstruction of the Lactobacillus population at subspecies level, we decided to perform an in-depth exploration of the ecological distribution of members of this genus in the collected cheese samples. Preliminary statistical analyses based on 16S rRNA gene microbial profiling data revealed that in cheese CST 1 the Lactobacillus genus displayed an average abundance of 32.69%, being universally present in all samples of cheese in CST 1 (Table S2; Fig. 2a and 3). Furthermore, the relative abundance of Lactobacillus was observed to be statistically different (Student t test, P < 0.01) between CST 1 and CST 2 (Table S4), thus corroborating the PCoA data (Table S2; Fig. 2a and 3).

Notably, we were unable to detect any statistically significant correlation between the relative abundance of lactobacilli and cheese type or specific cheese-making practices (Table 1; see also Table S4 in the supplemental material), suggesting that the composition of NWCs used is highly variable and that lactobacilli are well adapted to survive a wide range of environmental stresses induced by processing of milk for production of different cheeses. To support this statement, members of the Lactobacillus genus have been widely exploited as bacterial starters in cheese production due to their suitable features of acidification and flavor formation, their adaptability and survivability during cheese production, and their antimicrobial activity (18).

Assessment of the Lactobacillus population in Italian cheeses at the subspecies level.

All collected samples were subjected to lactobacillus ITS profiling, which is a recently developed microbiomic tool to determine the Lactobacillus composition in a given sample at the subspecies level (16). This approach generated a total of 100,892 sequence reads, with an average of 4,804 reads per sample (Table 1). In this context, cheese samples 5, 11, and 15 were excluded from further analysis due to their very low number of reads (Table 1). Intriguingly, cheese samples 5 and 11 also showed low Lactobacillus abundances determined using 16S rRNA gene-based profiling of 0.04 and 0.13%, respectively (Table S2).

Subsequently, ITS alpha-diversity analysis based on number of observed OTUs showed that all samples tend to reach a plateau, therefore indicating that the sequencing depth encompassed the majority of the biodiversity present in each sample (Fig. S1b). Notably, cheese 7 showed a high number of observed OTUs despite the low number of reads obtained (Table 1) and the low relative abundance of lactobacilli (1.2%) identified for this sample by 16S rRNA gene-based profiling (Table 1). The latter indicates high background noise in the obtained sequencing data and, for this reason, we decided to exclude the cheese 7 sample from further analysis.

Inspection of the biodiversity detected at 1,000 reads revealed that cheese produced from raw milk harbor a similar number of OTUs, ranging from 9 to 20 (average of 14 OTUs), with the single exception of cheese 21 (Emmentaler) that showed just three lactobacillus OTUs. In fact, the microbiota of cheese 21 was shown to be dominated by a single species, i.e., Lactobacillus helveticus, with a relative abundance of 99.84% (Table S5). Strains belonging to L. helveticus species have previously been observed to be dominant, and stable in in-house NWCs used for cheese making of Emmentaler and autochthonous strains of this taxon are considered pivotal for the development of the peculiar features of this cheese (17). Remarkably, no correlations were observed between abundance, biodiversity, and composition of the Lactobacillus population with cheese type or related metadata (Table 1).

A beta-diversity analysis was performed employing subspecies-level profiles obtained from Lactobacillus ITS profiling data and the Bray-Curtis index and then by means of a PCoA. Intriguingly, the PCoA revealed the presence of five clusters (ITS-1, ITS-2, ITS-3, ITS-4, and ITS-5) that represent five putative Lactobacillus community state types (LCSTs), numbered 1 to 5. Notably, while LCST 1 encompasses seven samples, the remaining clusters encompass just two or three samples (Fig. 2b), an observation indicative of less prevalent LCSTs. Thus, expansion of the analysis with a higher number of cheese samples is needed to increase the statistical power, while it may also reveal the presence of additional low-prevalence LCSTs with a distinct Lactobacillus composition.

Through an in-depth analysis of the Lactobacillus species composition obtained for the samples constituting the predicted clusters (Table S6; Fig. 4), we observed a codominance of L. helveticus and Lactobacillus delbrueckii subsp. bulgaricus in clusters LCST 1 and LCST 2, with average relative abundances of 57.23 and 31.07% in LCST 1 and 75.51 and 20.42% in LCST 2, respectively (Fig. 4b). Although LCSTs 1 and 2 seem quite similar in composition, all samples constituting LCST 1 showed a set of accessory microbial taxa, represented by subspecies of L. delbrueckii, that are absent in LCST 2 samples (Table S6; Fig. 4). These differences probably explain the obtained PERMANOVA P value of <0.01 when a statistical comparison was made between LCSTs 1 and 2 (Table S3).

FIG 4.

Taxonomic dissection of the Lactobacillus community in raw milk cheese. (a) Bar plot reporting the Lactobacillus species composition for each sample. (b) Average relative abundance of Lactobacillus ITS-based clusters.

Moreover, all other predicted LCSTs showed a high relative abundance of specific taxa. In this regard, LCST 3 is dominated by L. delbrueckii subsp. bulgaricus, with an average relative abundance of 40.69%, followed by L. delbrueckii subsp. jakobsenii at 19.96% and Lactobacillus salivarius at 23.73% (Fig. 4a and Table S6), LCST 4 is dominated by Lactobacillus plantarum, with an average relative abundance of 82.37%, while the dominant Lactobacillus species in LCST 5 is Lactobacillus paracasei, with an average relative abundance of 65.18% (Table S6 and Fig. 4). Intriguingly, the composition of the predicted LCST suggests that the raw milk and NWC (where used) microbiomes play a pivotal role in defining the composition of the bacterial population harbored by the corresponding cheeses. For example, six of seven samples contained in the LCST-1 cluster derive from buffalo milk, and all samples inside the LCST-3 cluster derive from goat milk, indicating a possible permanence of the Lactobacillus species community from original milk to the finished product. This is further validated by the finding that two cheeses (C6 and C16) cluster together in both 16S rRNA and Lactobacillus ITS microbial profiling-based PCoA, the same as four other cheeses (C14, C3, C2, and C1), suggesting that they contained a similar community at both bacterial genera and Lactobacillus species levels, This may be due to the finding that raw goat milk was used in the manufacturing process of cheeses C6 and C16, whereas raw buffalo milk was used in the manufacturing process of the other four cheeses, i.e., C14, C3, C2, and C1.

Furthermore, analysis of the prevalence of the detected species of lactobacilli revealed that, in addition to the cluster-defining taxa L. helveticus and L. delbrueckii subsp. bulgaricus, the species L. paracasei, L. plantarum, L. delbrueckii subsp. jakobsenii, and L. salivarius appear to be widespread across the analyzed cheeses, with an overall prevalence of >50% (see Table S7 in the supplemental material). Intriguingly, these Lactobacillus species are common bacterial starters used in cheese production due to their ability to remain stable during the production steps and because of their metabolic activities that are crucial for the formation of organoleptic characteristics of the final product (4). In this context, L. plantarum has also been shown to exert health-associated functions, thereby improving the nutritional and probiotic features of the final dairy products (19).

Notably, prevalence analysis also revealed that L. paracasei is present in all tested samples (Table S7). Intriguingly, L. paracasei has been reported to support the production of organoleptically stable cheese among production cycles, reducing colonization of unwanted bacterial genera and fungi inside cheese dough, primarily due to its high acidification rate (20). Thus, its presence is important for the improvement of the overall quality, safety, and consistency of many cheese types (20).

Data reported in the Web resource BacDive (21) was used to correlate LCSTs with optimal growth temperatures observed for the type strains of the Lactobacillus species detected by ITS profiling analysis (Table S8). Remarkably, we observed that the Lactobacillus species with an average relative abundance of >10% present in LCSTs 1, 2, 3, and 5 are able to optimally grow at 37°C, while the two taxa with an average relative abundance of >10% that represent LCST 4, i.e., L. plantarum and L. paracasei, can also efficiently grow at a temperature as low as 30°C (Table S8). These data indicate that the specific temperature conditions used by cheese makers for the coagulation of the curd play a key role in the selection of the microbiota harbored by cheese produced from raw milk.

Covariance analysis of lactobacilli in cheeses produced from raw milk.

Exploration of covariances between members of the genus Lactobacillus may provide insights into intragenus microbe-microbe interactions that underpin the ecology of lactobacilli in cheese produced from raw milk. For this reason, covariance analysis of the 20 Lactobacillus (sub)species detected by ITS profiling was performed using the Kendall index (22). Covariance data were then visualized through a force-driven network using the software Gephi. Intriguingly, modularity analysis revealed the presence of three main clusters of covariant lactobacilli (CCL), named according to the taxa with the highest overall average abundance, i.e., CCL-HB (L. helveticus, 40.0%; L. delbrueckii subsp. bulgaricus, 20.7%), CCL-PL (L. plantarum, 14.9%), and CCL-PA (L. paracasei, 14.5%) (Fig. 5) (Table S9). Notably, CCL-HB contains L. helveticus and L. delbrueckii subsp. bulgaricus, which are the key bacterial taxa defining PCoA clusters ITS-1 and ITS-2 (Fig. 2b and 4; see also Fig. S2 and Table S9 in the supplemental material). These data indicate that these lactobacilli tend to coexist in the same environment and suggest that they coevolved in the same ecological niche. Such microbe-microbe interactions contribute to the increased ecological fitness in the cheese environment and/or that the efforts of cheese producers to meet consumer requests in terms of desirable organoleptic qualities may have caused selection of the cooccurring taxa that we observed.

FIG 5.

Covariance of Lactobacillus species in cheese samples. The force-driven network was constructed using bacterial taxa as nodes and covariances as edges. Red edges correspond to negative correlations, while green edges represent positive associations. The node diameter is proportional to the average relative abundance of the corresponding Lactobacillus species.

Moreover, since all of these taxa have been found in cheeses produced with the use of NWCs, species constituting CCLs may be characterized by comparable performances in terms of thermal resistance and survivability in the whey. Furthermore, the minority microbial components of this cluster of cooccurring taxa are represented by other subspecies of L. delbrueckii (Fig. 5) that, intriguingly, are the accessory taxa shared by all the ITS-1 samples (Fig. 4). Remarkably, interactions with other clusters of lactobacilli are represented only by negative correlations (Fig. 5). In contrast, clusters CCL-PL and CCL-PA are characterized by a range of intra- and intercluster positive correlations (Fig. 5), which indicates that their cooccurrence may be the result of selection induced by industrial practices aimed at obtaining improved organoleptic features. Intriguingly, these data suggest that L. helveticus and L. delbrueckii subspecies specifically evolved to dominate in the cheese microbiota, and possibly in whey incubated for back slopping, at the expense of other lactobacilli that may have been present in the same ecological niche. This peculiar behavior explains why these two species are considered optimal microbial starters for cheese making, i.e., they are able to markedly modulate the cheese microbiota and standardize the organoleptic features of cheeses produced with raw and pasteurized milk, disregarding seasonal variation of the milk and the impact of the environmental microbiota. Notably, while a dominant behavior has been observed also for L. plantarum and L. paracasei, they seem to develop a network of positive correlations (indicating putative interactions) with other lactobacilli that may also play a role in defining the organoleptic features of cheese.

The composition of CCLs was also correlated with optimal growth temperatures previously reported for the type strains of each detected taxon (21) (Table S8). This analysis revealed that CCL-PL is represented by species whose type strains can efficiently grow at 30°C, with the only exception being Lactobacillus salivarius. Similarly, L. paracasei clusters with two other taxa that are able to efficiently grow at 30°C, thus supporting the notion that the microbiota of cheese produced from raw milk is shaped by the temperature used by cheese makers for coagulation of the curd.

Moreover, the relative abundance of the Lactobacillus genus observed by 16S rRNA gene profiling was shown to be correlated with the predicted abundance of CCL clusters, defined as the sum of its constituent taxa observed through ITS profiling. Our results highlight that CCL-HB is positively correlated with relative abundance of the Lactobacillus genus, while CCL-PL instead elicits a negative correlation (Table S10). Intriguingly, these results suggest that lactobacilli constituting CCL-HB and CCL-PL adapt to different CST of the whole cheese microbiota. In this context, we also observed that CCL-PL and CCL-PA showed a marked negative correlation with CCL-HB, confirming that the key taxa constituting CCL-HB, i.e., L. helveticus and L. delbrueckii subsp. bulgaricus, tend to dominate over other lactobacilli in the same ecological niche (Table S10).

Intriguingly, as described above, samples 6 and 16 display a similar 16S rRNA profile (Fig. 3), as well as Lactobacillus ITS rRNA patterns, sharing the presence of L. paracasei, L. salivarius, and L. plantarum (Fig. 4). However, the covariance network analysis reveals that only L. plantarum and L. salivarius cluster together, while L. paracasei fits in another cluster, which is not completely separated from the others (Fig. 5). This cross-reference may not universally apply to all cheeses and their bacterial communities, given their high level of complexity. Remarkably, since PCoA suggested the existence of low-prevalence LCSTs (see above), integration of this analysis with additional cheese samples may extend the network of observed interactions between cheese-colonizing lactobacilli.

Conclusions.

Multiple properties of specific cheese types, including texture, taste, flavor, and nutrient composition, are deeply influenced by the microbial population harbored by these dairy products (4). Despite its relevance for the cheese manufacturing industry, the microbial community present in cheeses has not yet been investigated in any detail by metagenomic approaches. The current knowledge of cheese microbiota composition is primarily based on cultivation-dependent methods that have been performed for a limited number of cheese types. Moreover, the relatively small number of microbiome-based studies focusing on cheese mainly relied on 16S rRNA gene profiling, which is a culture-independent approach allowing a precise taxonomic reconstruction, but only as far as the genus level (4). In order to obtain a comprehensive overview of the cheeses produced from raw milk that are widely consumed in Italy, 16S rRNA gene microbial profiling was used to reconstruct the genus-level composition of 21 Italian cheeses. Moreover, Lactobacillus ITS profiling was exploited to achieve an in-depth view of the (sub)species-level taxonomic composition of the Lactobacillus genus, a taxon of crucial relevance to cheese production.

The 16S rRNA sequencing data demonstrated the existence of distinct community state types, i.e., cheeses CST 1 and CST 2, characterized by dominance of Streptococcus/Lactobacillus or Lactococcus, respectively. Furthermore, in-depth analysis of the lactobacillus community composition at the (sub)species level revealed that certain Lactobacillus species tend to dominate or codominate in the same food matrix, allowing the clustering of collected samples into five groups representing putative Lactobacillus community state types (LCSTs). Interestingly, covariance analysis between the 20 Lactobacillus taxa detected in the cheese samples highlighted the cooccurrence of Lactobacillus species defining specific clusters of covariant lactobacilli (CCL). Collectively, these observations may be the starting point for the development of the rational optimized starters in order to improve the organoleptic features of final products. In fact, a correct balance between the LAB genera supports the species-level interactions responsible for both the resilience toward colonization by bacteria with negative effects on cheese maturation and the development of the expected organoleptic complexity, texture feel, and stability of the product (23). However, further ITS microbial profiling-based analyses aimed at dissecting the Streptococcus and Lactococcus populations typically present in cheeses are required to fully characterize the cheese microbiota since these two genera, together with Lactobacillus, define the most common CSTs found in cheese derived by fresh milk. In this context, the study of the existing species-level interactions between members of these dominant genera will provide drastic improvements in the formulation of next-generation starter cultures.

MATERIALS AND METHODS

Sample collection and bacterial DNA extraction.

A total of 21 Italian cheese samples produced from raw milk were collected from different cheese makers (Table 1). Details regarding the use of whey as the fermentation starter, production, and ripening can be found in Table 1. In this context, no precise data regarding temperature used for acidification is available since each cheese maker may choose a specific one, generally ranging between 35 and 45°C. All samples were kept on ice and shipped to the laboratory under frozen conditions where they were preserved at –80°C, until they were processed. Bacterial DNA extraction from the 21 cheese samples were performed using a DNeasy PowerFood microbial kit according to the manufacturer’s instructions (Qiagen, Germany).

16S rRNA/ITS microbial profiling.

Partial 16S rRNA gene sequences were amplified from extracted DNA using the primer pair Probio_Uni/Probio_Rev, which targets the V3 region of the 16S rRNA gene sequence (6). Partial Lactobacillus ITS sequences were amplified from extracted DNA by using the primer pair Probio-lac_Uni/Probio-lac_Rev, which targets the spacer region between the 16S rRNA and the 23S rRNA genes within the Lactobacillus rRNA locus (16). Illumina adapter overhang nucleotide sequences were added to the partial 16S rRNA gene-specific amplicons and to the generated ITS amplicons, which were further processed by utilizing the 16S metagenomic sequencing library preparation protocol (part 15044223 Rev. B; Illumina). Amplifications were carried out using a Verity thermocycler (Applied Biosystems). The integrity of the PCR amplicons was analyzed by electrophoresis on a Mupid-One electrophoresis system (Nippon Genetics, Japan). DNA products obtained following PCR-mediated amplification of the 16S rRNA gene and ITS sequences were purified by a magnetic purification step employing the Agencourt AMPure XP DNA purification beads (Beckman Coulter Genomics GmbH, Bernried, Germany) in order to remove primer dimers. The DNA concentration of the amplified sequence library was determined by using a fluorometric Qubit quantification system (Life Technologies). Amplicons were diluted to a concentration of 4 nM, and 5-μl quantities of each diluted DNA amplicon sample were mixed to prepare the pooled final library. Sequencing was performed using an Illumina MiSeq sequencer with MiSeq reagent kit (v3) chemicals.

16S rRNA/ITS microbial profiling analysis.

After sequencing, the fastq files were processed using QIIME2 software (24, 25). Paired-end reads were merged, and quality control retained sequences with a length between 140 and 400 bp, a mean sequence quality score of >25, and truncation of a sequence at the first base if a low-quality rolling 10-bp window was found. Sequences with mismatched forward and/or reverse primers were omitted. 16S rRNA gene and Lactobacillus ITS OTUs were defined at 100% sequence homology using DADA2 (26), and OTUs with fewer than two sequences in at least one sample were removed. All reads were classified to the lowest possible taxonomic rank using QIIME2 (24, 25) and a reference data set from the SILVA database (27), in the case of 16S rRNA gene sequences, or a custom Lactobacillus ITS database (16). The alpha-diversity was calculated based on the observed OTUs, while the beta-diversity was calculated using the Bray-Curtis dissimilarity.

Statistical analysis.

All statistical analyses, i.e., ANOVA, PERMANOVA, and the Student t test, as well as Kendall tau rank covariance analysis, were performed with SPSS software v. 22 (SPSS Statistics for Windows, v22.0; IBM Corp., Armonk, NY). The force-driven network was created using Gephi software (https://gephi.org/), and modularity was defined with a resolution of 0.6.

Data availability.

Raw sequences of 16S rRNA gene profiling together with Lactobacillus ITS profiling data are accessible through SRA study accession number PRJNA596087.