Shotgun Metagenomics and Volatilome Profile of the Microbiota of Fermented Sausages

ABSTRACT

Changes in the microbial gene content and abundance can be analyzed to detect shifts in the microbiota composition due to the use of a starter culture in the food fermentation process, with the consequent shift of key metabolic pathways directly connected with product acceptance. Meat fermentation is a complex process involving microbes that metabolize the main components in meat. The breakdown of carbohydrates, proteins, and lipids can lead to the formation of volatile organic compounds (VOCs) that can drastically affect the organoleptic characteristics of the final products. The present meta-analysis, performed with the shotgun DNA metagenomic approach, focuses on studying the microbiota and its gene content in an Italian fermented sausage produced by using a commercial starter culture (a mix of Lactobacillus sakei and Staphylococcus xylosus), with the aim to discover the connections between the microbiota, microbiome, and the release of volatile metabolites during ripening. The inoculated fermentation with the starter culture limited the development of Enterobacteriaceae and reduced the microbial diversity compared to that from spontaneous fermentation. KEGG database genes associated with the reduction of acetaldehyde to ethanol (EC 1.1.1.1), acetyl phosphate to acetate (EC 2.7.2.1), and 2,3-butanediol to acetoin (EC 1.1.1.4) were most abundant in inoculated samples (I) compared to those in spontaneous fermentation samples (S). The volatilome profiles were highly consistent with the abundance of the genes; elevated acetic acid (1,173.85 μg/kg), ethyl acetate (251.58 μg/kg), and acetoin (1,100.19 μg/kg) were observed in the presence of the starters at the end of fermentation. Significant differences were found in the liking of samples based on flavor and odor, suggesting a higher preference by consumers for the spontaneous fermentation samples. Inoculated samples exhibited the lowest scores for the liking data, which were clearly associated with the highest concentration of acetic acid.

IMPORTANCE We present an advance in the understanding of meat fermentation by coupling DNA sequencing metagenomics and metabolomics approaches to describe the microbial function during this process. Very few studies using this global approach have been dedicated to food, and none have examined sausage fermentation, underlying the originality of the study. The starter culture drastically affected the organoleptic properties of the products. This finding underlines the importance of starter culture selection that takes into consideration the functional characteristics of the microorganism to optimize production efficiency and product quality.

INTRODUCTION

During meat fermentation, innumerable biochemical reactions take place that involve the breakdown of the main meat components (proteins, carbohydrates, and lipids) and the consequent conversion of metabolites that, for the most part, have an important organoleptic impact (1). In the last decades, several culture-dependent and -independent studies were carried out to describe the evolution of the microbiota in fermented meat and meat products (2–7). By using recent 16S amplicon target sequencing, it was found that meat can be contaminated by several microbial groups, with spoilage potential coming from water, air, and soil, as well as from the workers and the equipment involved in the processing or during the slaughtering procedures (8). Recently, it was assessed that more than 30 different genera of Staphylococcaceae and Lactobacillaceae are present along with several contaminant species during the fermentation and ripening of fermented sausages (9), but only a few of these taxa were recognized as metabolically active (10). More specifically, the predominance of Lactobacillus sakei during spontaneous fermentation was shown, usually associated with the presence of members of Leuconostocaceae or with other bacteria (Lactobacillus curvatus, Leuconostoc carnosum, Staphylococcus xylosus, and Staphylococcus succinus) (10). The development of coagulase-negative cocci (CNC) during meat fermentation can contribute to the proteolysis and lipolysis of meat components, while the lactic acid bacteria are responsible for the rapid decrease of pH, the production of lactic acid, and the production of small amounts of acetic acid, ethanol, acetoin, carbon dioxide and pyruvic acid (6, 9).

It was recently shown that the perturbation of the food system due to different ripening conditions (11), the presence of starter culture, and changes in the quality of the raw materials (initial microbiota composition) can change the genetic repertoire of the microbes. Changes in the composition of the microbiota have an impact on the sensorial characteristics of a product. Such changes may be due to differences in the abundances of genes encoding enzymes that are involved in biochemical reactions leading to volatile compounds (11).

The evolution of massive sequencing technologies such as shotgun DNA sequencing (DNA-seq) or RNA sequencing (RNA-seq) can help researchers understand and characterize the microbial composition and function of the microbiota in a food ecosystem. Unlike RNA-seq, which only enables profiling of the transcriptome, shotgun DNA-seq offers the opportunity to concomitantly perform compositional analyses of the existing microbiota and gene pool (12). This technique is used largely in several environmental systems (human or agriculture) and has only been applied in a few studies of food to discover the presence of pathogens (13) and toxins (14) or to discover the gene content during food processing in vegetables (kefir grain [15], kimchi [16], and soy [17]), broiler meat (18), and cheese (19).

The present study used metagenomic DNA-seq analyses to examine the microbiota composition and the gene content during the ripening of the traditional Italian Felino sausages and to investigate how the use of starter culture can affect the bacterial gene abundance and the volatilome profile of the sausages.

RESULTS AND DISCUSSION

Microbial community profile.

Microbial counts on De Man Rogosa Sharpe (MRS) agar and mannitol salt agar (MSA) clearly showed the increase (P < 0.05) in the density of the lactic acid bacteria and Staphylococcaceae populations in the inoculated meat (I) samples compared to that in the spontaneous fermentation (S) samples. A significant reduction in the Enterobacteriaceae population (P < 0.05) was also observed early during fermentation in the I samples (see Table S1 in the supplemental material), likely due to the ability of the starter culture to lower the pH faster (S3 versus I3, P < 0.05) (Table S1) and to compete with the indigenous microbiota.

Metagenomic shotgun sequencing data were used to estimate the relative abundance of microbial cells by mapping the nonhost clean reads against a set of clade-specific marker sequences by using MetaPhlAn2, which enables the estimation of relative abundance for individual species (20). Taxonomical assignment in I samples showed the dominance of L. sakei and S. xylosus followed in reduced proportion by L. curvatus, Staphylococcus equorum, and Acinetobacter sp. (Fig. 1).

FIG 1.

Taxonomy analysis of fermented sausages. The plot shows the distribution of taxa during ripening. Only taxa with an incidence above 0.5% in at least 2 samples are shown. Samples are labeled according to time (0, 3, 7, and 40 days) and type (S, spontaneous; I, inoculated).

The spontaneous fermentation samples showed the presence of L. sakei (varied from 37 to 56% relative abundance) and L. curvatus (10 to 20%) among the most abundant taxa, followed by a number of minor taxa identified as S. xylosus, Leuconostoc sp., Lactococcus garvieae, and Lactococcus lactis, as well as Acinetobacter, Pseudomonas, and Propionibacterium.

The isolation from agar plates, identification, and fingerprinting of Lactobacillaceae and Lactococcaceae (LAB) and staphylococci showed the presence of L. sakei and S. xylosus in 90% of the colonies purified from the plates of both I and S samples, with few strains belonging to Weissella hellenica, Enterococcus faecalis, and Kocuria sp. The repetitive extragenic palindromic (REP) fingerprint analysis showed the ability of the starters to dominate the LAB and staphylococcal microbiota from the beginning, whereas several REP biotypes of L. sakei and S. xylosus were found in the S samples (data not shown). The higher presence of sequences belonging to L. sakei and the finding of several REP biotypes in the spontaneous fermentation samples clearly indicated that fermented meat is an ecological niche for L. sakei. The several strains of L. sakei present in the spontaneous fermentation samples most probably originated from the animal (21) and/or were introduced during food processing (10).

Sequencing and assembly of the sausage metagenomes.

A total of 29.47 Gbp of raw reads was generated from 16 samples, which yielded 1.84 Gbp/sample. The quality reads after trimming were 22.04 Gbp of sequences. After host sequence removal, 8.59 Gbp of clean reads was analyzed, and for each sample, approximately ∼2 Mbp to ∼6 Mbp of reads was obtained (see Table S2). Rarefaction curves obtained by using MEGAN were used to determine gene abundance richness. These revealed that bacterial diversity was well represented, as they are nearly parallel with the x axis (see Fig. S1), although the inoculated samples showed a lower gene abundance than the spontaneous fermentation samples, in particular at time zero.

Sausage samples displayed higher proportions of reads of host origin (ranging from 43 to 87%), especially at the beginning of the fermentation. This is due to the higher abundance of mammalian cells and the low microbial biomass, especially at the beginning of the ripening. Similar results have already been displayed in other food matrixes (18, 22).

A de novo-performed assembly generated a total of 9,755 contigs of more than 1,000 bp in length, with an average N50 of 1,587 bp for spontaneous fermentation and 83,842 bp for the inoculated ones (Table S2). Consistent with the reduced microbiota diversity in I samples (Fig. 1), there were significantly fewer total genes predicted and higher N50 values in the assembled metagenomic data of I samples than in S samples.

Exploration of metabolic potential of the sausage microbiota.

To explore the metabolic potential, we classified the predicted genes by aligning them to the integrated NCBI-NR database of nonredundant protein sequences. A total of 11,402 predicted genes were identified, of which 10% were assigned to KEGG pathways by MEGAN, which compares genes using blastx and then assigns them to the latest common ancestors (LCA) of the targeted organisms. The KEGG analysis assigned 1,774 genes to 21 pathways (Fig. 2), and the results gave a highly integrated picture of the global sausage microbiota metabolism. Consider the KEGG annotations at level 2, in which the KEGG categories carbohydrate metabolism, amino acid metabolism, translation, and nucleotide metabolism (Fig. 2) were found as the most abundant throughout the ripening period, consistent with those observed on a cheese matrix using a similar approach (19, 23). However, the genes for carbohydrate and lipid metabolism increased significantly (P < 0.05) only in spontaneous fermentation sausages during ripening.

FIG 2.

Functional classification of fermented sausages during ripening. Functional classes were determined according to the first level of the KEGG annotations. Samples are color coded according to time (0, 3, 7, and 40 days) and type (S, spontaneous; I, inoculated). Data in the two batches for each sampling time were averaged.

Within the translation and nucleotide metabolism categories, the genes related to ribosomal protein and DNA polymerase, respectively, were the most abundant in all the data sets. The higher abundance of genes related to nucleotide metabolism is a consequence of the sugar consumption in meat during ripening and the nucleoside metabolism that might improve LAB survival on meat during ripening after sugar consumption (24).

Taking into account the KEGG pathways related to carbohydrate, amino acid, and lipid categories, the pathway network (Fig. 3) shows the evolution of those pathways over time and under both conditions. Sample node size is proportional to a pathway's abundance in a given sample. In particular, from the thickness of the edges, it was possible to visualize the relative abundance of numerous genes encoding proteins with functions related to energy metabolism, including all enzymes in the pathways related to glycolysis/gluconeogenesis, the pentose phosphate pathway, fructose and mannose metabolism, and amino sugar and nucleotide sugar metabolism. These were most abundant after 3 days of fermentation in the inoculated samples. On the other hand, the same pathways in the S samples were found to be most abundant after 7 days of fermentation, most likely due to the delayed evolution of the microbiota (lactic acid bacteria and Staphylococcaceae) relative to that in the inoculated samples (Fig. 3).

FIG 3.

Relationships between metabolic pathways and samples. KEGG network summarizing the relationships between metabolic pathways related to carbohydrates (red), amino acids (yellow), and lipids (blue) and samples (cyan, spontaneous; green, inoculated). Metabolic pathways and samples are connected with lines (i.e., edges) for which the thickness is proportional to the abundance of that pathway in the connected sample.

The KEGG gene differences between S and I samples were further observed by using the principal-component analysis (PCA; see Fig. S2). The PCA showed a separation of the metagenome contents between spontaneous fermentation and inoculated samples. Genes involved in glycolysis were the most abundant, and in particular, acetaldehyde dehydrogenase had the highest number of sequence assignments in the entire data set, followed by alcohol dehydrogenase, enolase, acetate kinase, phosphoketolase, and gluconokinase.

Metagenomic content boosted the production of VOCs during ripening.

Going more deeply into the metagenome content, the DESeq2 analysis identified 340 KEGG genes differentially abundant between spontaneous fermented sausages and inoculated ones (false-discovery rate [FDR], <0.05) (see Table S3). According to the DESeq2 analysis, the most prominent differences during fermentation among the two sausage types involved key KEGG genes in carbohydrate metabolism (pyruvate metabolism and glycolysis).

KEGG genes responsible for the reduction of acetaldehyde to ethanol (EC 1.1.1.1), acetyl phosphate (acetyl-P) to acetate (EC 2.7.2.1), and 2,3-butanediol to acetoin (EC 1.1.1.4) were most abundant in I samples compared to those in S samples (see Fig. S3a). Additionally, the elevated production of acetyl-P, acetaldehyde, and 2,3-butanediol increased acetic acid (1,173.85 μg/kg), ethyl acetate (251.58 μg/kg), and acetoin (1,100.19 μg/kg) during fermentation in the presence of starter cultures (P < 0.05) as observed from GC analysis (Fig. 4 and Table S4).

FIG 4.

Abundance of VOCs during ripening. Acetic acid, acetoin, and ethyl acetate concentrations over time (0, 3, 7, and 40 days) and under two fermentation conditions (red, inoculated; blue, spontaneous fermentation). Boxes represent the interquartile ranges (IQRs) between the first and third quartiles, and the lines inside represent the medians (2nd quartiles). Whiskers denote the lowest and the highest values within IQRs from the first and third quartiles, respectively. Circles represent outliers beyond the whiskers.

Acetoin is the ketone found mostly in fresh meat stored under different conditions and is regarded a product of the carbohydrate catabolism of LAB and staphylococci associated with the cheesy odors of meat (25, 26). Moreover, even though we found a high presence of this molecule at the end of ripening in I samples (1,100 μg/kg, Table S4), Dainty et al. (27) have reported that the higher presence of acetoin in meat is not unpleasant. Carbohydrate metabolism was the most prevalent pathway in I samples, most likely due to the predominant presence of the heterofermentative L. sakei (28), and carbohydrate pathways were found to be one of the main precursors of many volatile organic compounds (VOCs) such as acetate, acetoin, diacetyl, acetic acid, and isobutyric acid (1).

Heterofermentative carbohydrate metabolism (alternative degradation of pyruvate) in the starter culture is well explained (29), and it was found that inoculated fermentation results in the highest abundance of acetate kinase (EC 2.7.2.1), which can lead to the formation of acetic acid, a typical aroma compound of dry fermented sausages (30). L. sakei utilizes glucose and fructose, as well as several hexoses and amino acids, as primary energy sources during the initial growth stage (31). In particular, sugars are fermented through different metabolic pathways: sugar hexose fermentation is homolactic and proceeds via the glycolytic pathway leading to lactate, whereas pentoses are fermented through the heterolactic phosphoketolase pathway, ending with lactate and other end products such as acetate (32).

Glucose is the preferred carbon source for L. sakei (24) in meat in chilled storage, and after its utilization, several substrates are metabolized, such as lactate, gluconate, glucose-6-phospate, pyruvate, propionate, formate, ethanol, acetate, amino acids, nucleotides, etc. (1).

Regarding the evolution of VOCs during ripening (Fig. 5), we observed that I3 and I7 cluster together with S7 and display significantly higher abundances (P < 0.05) of short-chain esters, such as ethyl acetate, ethyl 2-methylbutanoate, ethyl isovalerate, and ethyl butanoate, and some short-chain fatty acids (SCFA; acetic acid and butanoic acid) (Fig. 5 and Table S4). Samples at the end of ripening (I40 and S40) cluster together; however, I40 displays significantly higher abundances (P < 0.05) of ethyl alpha-hydroxybutyrate, ethyl ester, 3-methyl-2-buten-1-ol, and acetoin, while S40 displays a higher presence of ethyl decanoate and 2-heptanol (P < 0.05) (Fig. 5 and Table S4). This is in agreement with the finding that the inoculated samples boosted the development of microorganisms that can induce increased formation of ethanol and acetic acid at the beginning of ripening (P < 0.05). In addition, the most abundant acetate kinase (EC 2.7.2.1) in I samples (false-discovery rate [FDR], <0.05) may be involved in the interconversion of 2-oxobutanoate to propanoate from amino acid metabolism (serine and aspartate) that can lead to the production of short-chain volatile esters that were most abundant in I samples compared to that in S samples (Fig. S3b). In detail, it was shown that L. sakei is auxotrophic for all amino acids except aspartic and glutamic acids and needs to absorb them after amino acid metabolism (33). Meat provides an environment rich in amino acids, and L. sakei was shown to be able to use it as an energy source. In particular, inosine metabolism can lead to the formation of acetic acid as well as ethanol (34).

FIG 5.

Correlation patterns between VOCs and samples. Correlations between the abundance of VOCs and spontaneous fermented (S) and inoculated (I) samples. Rows and columns are clustered by Ward linkage hierarchical clustering. The intensity of the colors represents the degree of correlation between the samples and VOCs as measured by Spearman's correlations.

The indigenous microbiota of the S samples (L. lactis, Leuconostoc citreum, Leuconostoc gelidum, S. xylosus, and L. sakei) displayed higher counts of KEGG genes involved in ex novo fatty acids biosynthesis from pyruvate and amino acid metabolism (see Fig. S4). Consistent with this and associated with the fruity and sweet odor description, long-chain esters such as ethyl octanoate and decanoate were more abundant in S samples than in I samples at the end of ripening (see Table S4) (P < 0.05).

The differential abundance analysis showed that S samples displayed a high abundance of KEGG genes compared to that of I samples (Table S3). In particular, it was possible to find several genes encoding proteases and amino acid catabolism, but no clear association could be found between the genes and the volatilome profiles. We also observed unusual functions of L. sakei. Specifically, key genes in folate synthesis, such as genes for dihydrofolate synthase/folylpolyglutamate synthase (EC 6.3.2.12 and EC 6.3.2.17) and formate tetrahydrofolate ligase (EC 6.3.4.3), associated with the indigenous presence of L. sakei, were found in S samples. The ability of several strains of L. sakei to exhibit genes related to folate biosynthesis was assessed elsewhere (16). In addition, nucleotide metabolism serves as one of the important sources of energy for L. sakei. As recently pointed out (24), this alternative energy route can be explained by the presence of functions unusual for LAB due to a methylglyoxal synthase-encoding gene (EC 4.2.3.3) that, in this study, we observed was more abundant in I samples than in S samples (Table S4) (FDR < 0.05).

Correlations between metagenome, volatilome, and sensorial characteristics.

Spearman's correlations (false-discovery rate [FDR], <0.05) between KEGG pathways, taxa, and the volatilome (Fig. 6) clearly suggest that L. sakei was positively correlated with several pathways involved in carbohydrate and amino acid metabolism, namely, amino sugar, fructose, glycolysis, and pentose phosphate pathways, and valine, leucine, and isoleucine degradation, as well as with the abundance of acetoin, ethyl 2-methylbutanoate, and 3-methyl-2-buten-1-ol. Branched-chain esters derived from amino acids such as isoleucine and leucine are precursors of the important aroma compounds (branched-chain alcohols, aldehydes, acids, and their corresponding esters), and the majority of branched-chain flavor compounds in sausages are usually generated at the end of ripening and after the growth of the staphylococci has ceased (35, 36) due to the proteolytic activity of LAB that absorb nutrients after sugar consumption by hydrolyzing proteins present in the environment (37). In agreement with this, we also found several associations between LAB and ester compounds. The presence of L. lactis positively correlated with galactose and butanoate metabolism as well as with 2-3-octanedione and ethyl-alpha-hydroxybutyrate (Fig. 6), L. brevis correlated with ethyl esters, L. citreum correlated with ethyl isovalerate, while Leuconostoc sp. displayed the highest negative correlation with the volatile esters (ethyl butanoate, octanoate, and pentanoate) as well as with fatty acid metabolism. However, esters are also formed through the esterification of alcohols (ethanol in particular) and carboxylic acids found in meat (38) following microbial esterase activity (30, 39).

FIG 6.

Correlations between taxa, ripening-related metabolic pathways, and volatilome data. Correlation network showing significant (false-discovery rate [FDR], <0.05) Spearman's correlations between KEGG genes belonging to amino acid and lipid metabolism, VOCs, and taxa. Node sizes are proportional to the numbers of significant correlations. Colors of the edges indicate negative (blue) or positive (pink) correlations.

The higher loads of the LAB and the finding of a higher presence of L. sakei in inoculated fermentation, along with the higher concentrations of volatile metabolites (in particular, of the acids class), enabled a definite separation of the samples according to the presence of the starter culture. Significant differences were found among samples in terms of liking on the basis of flavor and odor (P < 0.05) (Fig. 7), suggesting a higher preference by consumers for the S samples. The radar plots depicted in Fig. 7 clearly show that I samples exhibited the lowest scores for the liking data, which were clearly associated with the highest presence of acetic acid and its related odor descriptors, namely, pungent, acidic, cheesy, and vinegar (1).

FIG 7.

Liking test. (A) Radar graphs displaying the liking of appearance, odor, taste, flavor, and texture and overall liking expressed by consumers for the sausages made by spontaneous and inoculated fermentation. (B) Distributions of the liking scores of flavor and odor (P < 0.05) for fermentation conditions (red, inoculated; blue, spontaneous fermentation). Boxes represent the interquartile ranges (IQRs) between the first and third quartiles, and the lines inside represent the medians (2nd quartiles). Whiskers denote the lowest and the highest values within IQRs from the first and third quartiles, respectively. Circles represent outliers beyond the whiskers.

Conclusion.

In this study, we present an integrated analysis of a typical Italian fermented sausage with a strict link between the volatilome profile, microbiota, gene content, and consumer acceptability. A robust standard bioinformatics pipeline to process, annotate, and realize the sausages' gene catalog was assessed in accordance with several pipelines in different environments. We found that the presence of the starter culture, in particular, the presence of L. sakei, ensured a fast and predominant growth, high acidification rate, and fast consumption of fermentable substrates. A decline in the numbers of Enterobacteriaceae was observed, as well as a decline in microbial diversity. In addition, the pH endpoint of sausages made with the addition of starter cultures was lower than the pH of those made by spontaneous fermentation. On the other hand, the starter cultures used in this study had a negative impact on the sensory properties of the product, as confirmed by the consumer test, due to the faster metabolic activity implied by the metagenomics data and confirmed by the meta-metabolomics data. Spontaneous fermented sausages made without the addition of starter cultures were generally more acceptable and displayed a higher variety of genes with valuable potential (such as those involved in folate biosynthesis).

The multiomics approach followed, in this case DNA-seq metagenomics coupled with metabolomics data, was effective in providing unprecedented insight into fermentation mechanisms that can affect the final characteristics of products.

MATERIALS AND METHODS

Sausage manufacturing and sample collection.

Felino-type sausages were manufactured in a local meat factory in the area of Turin according to the standard recipe. The formulation used in the manufacture included pork meat (77%), lard (23%), salt (2.9%), spices (including pepper, coriander, nutmeg, and cinnamon [0.2%]), sucrose (0.4% [wt/wt]), nitrate salt (E252; 0.01%), and wine (0.3%). A commercial starter culture composed of Lactobacillus sakei and Staphylococcus xylosus (SA8-400M; Veneto Agricoltura, Thiene, Vicenza, Italy) was added to the meat batter to reach the final concentration of 5 log CFU/g. The meat batter was stuffed into synthetic casings, resulting in 12 sausages of about 5 cm in diameter and 500 g in weight. Fermentation and ripening were carried out in a climatic chamber; time and relative humidity/temperature conditions are reported in Table S5 in the supplemental material. Another series of 12 sausages was prepared as described above, without using the starter culture, and used as a control. Three samples of the meat mixture prior to filling (0) and three sausage samples obtained after 3, 7, and 40 days of fermentation/ripening were analyzed.

Two independent batches were analyzed for a total of 24 inoculated (I) and 24 spontaneously fermented (S) sausages. Both batches were prepared with meat from the same meat factory at two different periods of time, the second batch 1 week after the first. At each sampling point, 3 sausages for both conditions (I and S) were removed from the casings and individually placed in a stomacher bag (Sto-circul-bag; PBI, Milan, Italy) and gently mixed. Aliquots were then used for microbial count, pH and a_w determination, DNA extraction, and volatile organic compound (VOC) analysis.

Microbiological analysis and pH and a_w determination.

Approximately 25 g from each of the three sausages at every sampling time was homogenized with 225 ml of Ringer's solution (Oxoid, Milan, Italy) for 2 min in a stomacher (LAB Blender 400; PBI, Italy). Decimal dilutions in quarter-strength Ringer's solution were prepared, and aliquots of 0.1 ml of the appropriate dilutions were spread in triplicates on the following media: (i) gelatin peptone agar (GPA; Oxoid) for total aerobic bacteria incubated for 48 to 72 h at 30°C; (ii) De Man Rogosa and Sharpe agar (MRS; Oxoid) for Lactobacillaceae and Lactococcaceae (LAB) incubated at 30°C for 48 h; (iii) mannitol salt agar (MSA; Oxoid) for Staphylococcaceae incubated at 30°C for 48 h; and (iv) violet red bile agar (VRBA; Oxoid) for Enterobacteriaceae incubated at 30°C for 24 to 48 h. The results were calculated as the means of log CFU from three independent determinations. The pH was measured by immersing the pH probe of a digital pH meter (micropH2001; Crison, Barcelona, Spain) in a diluted and homogenized sample containing 10 g of sausages and 90 ml of distilled water. Water activity (a_w) was measured with a calibrated electric hygrometer (HygroLab; Rotronic, Bassersdorf, Switzerland) according to the manufacturer's instructions. Fifteen colonies from MRS agar and MSA at each sampling point were randomly isolated and purified. The purified isolates were preliminarily characterized by Gram staining and microscopic observations, as well as by catalase and oxidase reactions. Working cultures were maintained in brain heart infusion (BHI; Oxoid) or MRS (Oxoid) broth with 25% glycerol and stored at −20°C.

Molecular typing by rep-PCR and RSA and cluster analysis and identification.

LAB and Staphylococcus sp. isolates were subjected to DNA extraction as previously reported (10). The molecular identification of the LAB isolates was performed by PCR 16S-23S rRNA gene spacer analysis (RSA) and 16S rRNA gene sequencing. The RSA was carried out with primers G1 (GAAGTCGTAACAAGG) and L1 (CAAGGCATCCACCGT) under conditions reported by Bautista-Gallego et al. (40). LAB isolates displaying the same RSA profiles were then subjected to identification. LAB and Staphylococcaceae molecular fingerprints were obtained by using repetitive extragenic palindromic PCR (rep-PCR) with the primer (GTG)5 according to Iacumin et al. (41). The rep-PCR profiles were normalized, and cluster analysis was performed using Bionumerics software (version 6.1; Applied Maths, Sint-Martens-Latem, Belgium). The dendrograms were calculated on the basis of the Dice coefficient of similarity with the unweighted pair group method using arithmetic averages (UPGMA) clustering algorithm (42). For Staphylococcus, after cluster analysis, 2 isolates from each cluster at 70% similarity were selected and subjected to identification. The identification of LAB and Staphylococcaceae was performed by amplifying the 16S rRNA gene. The oligonucleotide primers described by Weisburg et al. (43), FD1 (5′-AGA GTT TGA TCC TGG CTC AG-3′) and RD1 (5′-AAG GAG GTG ATC CAG CC-3′) (Escherichia coli positions 8 to 17 and 1540 to 1524, respectively), were used. PCR conditions were chosen according to Ercolini et al. (44). 16S rRNA amplicons were sent for sequencing to GATC-Biotech (Cologne, Germany). To determine the closest known relatives of the 16S rRNA gene sequences obtained, searches were performed in public data libraries (GenBank) with the BLAST search program.

Analysis of volatile organic compounds.

The volatile organic compounds (VOCs) in the sausage samples were extracted using headspace (HS) solid-phase microextraction (SPME) and analyzed by gas chromatography-mass spectrometry (GC/MS). All samples were analyzed in triplicates. The analysis was conducted using a 20-ml vial filled with 3 g of sample to which 10 μl of 3-octanol in ultrapure water (323 ppb) and methyl caproate (3,383 ppb) were added as internal standards for ester chemical class. After an equilibration time of 5 min at 40°C, the extraction was performed using the same temperature for 30 min with a 50/30 μm DVB/CAR/PDMS fiber (Supelco, Milan, Italy) with stirring (250 rpm) using an SPME autosampler (PAL System; CombiPAL, Switzerland). The fiber was desorbed at 260 for 1 min in splitless mode. GC/MS analysis was performed with a GC-2010 gas chromatograph equipped with a QP-2010 Plus quadruple mass spectrometer (Shimadzu Corporation, Kyoto, Japan) and a DB-WAXETR capillary column (30 m by 0.25 mm, 0.25-μm film thickness; J&W Scientific Inc., Folsom, CA). The carrier gas (He) flow rate was 1 ml/min. The temperature program began at 40°C for 5 min, and then the temperature was increased at a rate of 10°C/min to 80°C and 5°C/min to 240°C for 5 min. The injection port temperature was 260°C, the ion source temperature was 240°C, and the interface temperature was 240°C. The detection was carried out by electron impact mass spectrometry in total ion current mode, using an ionization energy of 70 eV. The acquisition range was m/z 33 to 330. The identification of volatile compounds was confirmed by the injection of pure standards and the comparison of their retention indices (a mixture of a homologous series of C5 to C28 was used) with MS data reported in the literature and in the database found at http://webbook.nist.gov/chemistry/. Compounds for which pure standards were not available were identified on the basis of mass spectra and retention indices available in the literature. Semiquantitative data (μg/kg) were obtained by measuring the relative peak area of each identified compound in relation to that of the added internal standard.

DNA extraction, library preparation, and sequencing.

At each sampling point, 2 ml of the first 10-fold serial dilution was collected and directly centrifuged at maximum speed for 30 s. Total DNA was extracted from the pellet by using the MasterPure complete DNA and RNA purification kit (Illumina Inc., San Diego, CA) according to the manufacturer's instructions. Three biological replicates (3 different sausages at each sampling point and for each condition) were subjected to DNA extraction, and total DNA was pooled before further processing. The DNA was further purified using the Agencourt AMPure XP (Beckman Coulter, USA) according to the manufacturer's protocol. DNA quality was checked on the NanoDrop 2000c instrument (Thermo Scientific, USA) and quantified on a Qubit 2.0 fluorometer (Invitrogen, USA). Sequence libraries were fragmented and tagged with sequencing adapters by using the Nextera XT library preparation kit (Illumina) according to the manufacturer's instructions. The libraries were quantified using a Qubit 2.0 fluorometer. The quality and the size distribution of the libraries were determined by using high-sensitivity DNA chips and DNA reagents on a Bioanalyzer 2100 (Agilent, USA). Sequencing was performed in the MiSeq (Illumina) system for a 151- cycle paired-end run by sequencing 4 samples/run. Base calling and Illumina barcode demultiplexing processes were performed by the MiSeq control software v2.3.0.3, the RTA v1.18.42.0, and the CASAVA v1.8.2.

Bioinformatics and data analysis.

Raw read quality (Phred scores) was evaluated by using the FastQC toolkit (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). The raw sequences were trimmed to a Q score of 30 with SolexaQA++ software (45). Duplicate sequences and sequences of less than 50 bp were discarded by using Prinseq (46). The obtained trimmed reads were then mapped to the draft genome of Sus scrofa (ftp://ftp.ncbi.nlm.nih.gov/genomes/Sus_scrofa/Assembled_chromosomes/) by using Bowtie2 (47) in end-to-end sensitive mode. Clean nonhost reads were then assembled with Velvet (48), with a minimum contig length set at 600 bp. Each contig was run through an automated gene annotation pipeline utilizing MetaGeneMark (49). Predicted genes for all the samples were concatenated and clustered using USEARCH (50) with the following criterion: identity of ≥95% and an alignment length of >90%. The sausages' gene catalogs obtained were then aligned against the NCBI-NR database by using mblastx (51) to obtain the gene annotation. Clean reads were mapped back by using Bowtie2 to the annotated gene catalogue to enable semiquantitative analysis and to check the quality of the assembly. The number of reads uniquely mapped to each gene (SAM file) were then used for functional analysis based on the Kyoto encyclopedia of genes and genomes (KEGG) (52) version April 2011 using MEGAN software version 5 (53). MEGAN specifically requires BLAST searches of nucleotide or protein sequences as input. The software then parses the BLAST results using the NCBI Refseq identification numbers (IDs) to assign peptides to KEGG pathways. The KEGG classification in MEGAN is represented by a rooted tree (with approximately 13,000 nodes) whose leaves represent different pathways.

The functional gene count table was internally normalized in MEGAN by checking the “use normalized count” option. Rarefaction analysis was performed based on the leaves of the tree in MEGAN. The phylogenetic characterization of the shotgun sequences was achieved at the species level of taxonomy by using MetaPhlAn2 (20) with default parameters. Statistical analysis and plotting were carried out in an R environment. Data normalization and determination of differentially abundant genes were then conducted using the Bioconductor DESeq2 package (54) in the statistical environment R. P values were adjusted for multiple testing using the Benjamini-Hochberg procedure, which assesses the false-discovery rate (FDR).

Pairwise Spearman's correlations between taxa, KEGG genes, and volatile organic compounds were assessed by the R package psych, and the significant correlations (FDR, <0.05) were plotted in a correlative network by using Cytoscape v. 2.8.143. A principal-component analysis (PCA) and hierarchical clustering were carried out by using the made4 package in R. All the results are reported as mean values from the two batches.

Liking test.

To assess the sensory acceptability of sausage samples at the end of the ripening, a consumer test was performed. A total of 15 regular consumers of sausages (7 male and 8 female participants; age, 28 to 56 years) voluntarily participated in the sensory evaluation. Sausage samples (10 g) were served under blind conditions in opaque white plastic cups coded with a random three-digit number. Samples were served in a completely randomized order, with the spontaneously fermented sausages served as the last sample for all subjects to limit the contrast effect. Consumers were asked to observe its appearance, smell, and taste and rate the sausages for appearance, odor, taste, flavor, texture, and overall acceptance. Liking was expressed on a 9-point hedonic scale ranging from “dislike extremely” (1) to “like extremely” (9). Purchase interest (i.e., would you buy this sausage?) was also rated on a 7-point scale (1, absolutely no; 7, absolutely yes). Participants were required to rinse their mouths with tap water for approximately 1 min between samples. Consumers took between 15 and 20 min to complete the evaluation. Liking data (appearance, odor, taste, flavor, texture, and overall acceptance) and declared purchase interest from consumers were independently subjected to a pairwise Kruskal-Wallis test in the R environment assuming samples as the main factors.

Accession number(s).

The raw read data were deposited in the Sequence Read Archive of NCBI (accession number SRP092525).