Longitudinal Metatranscriptomic Analysis of a Meat Spoilage Microbiome Detects Abundant Continued Fermentation and Environmental Stress Responses during Shelf Life and Beyond

It is generally known which organisms are the typical SSO in foods, whereas the actively transcribed genes and pathways during microbial succession are poorly understood. This knowledge is important, since better approaches to food quality evaluation and shelf life determination are needed. Therefore, we conducted this study to find longitudinal markers that are connected to quality deterioration in a MAP beef product. This kind of RNA marker could be used to develop novel types of rapid quality analysis tools in the future. New tools are needed, since even though SSO can be detected and their concentrations determined using the current microbiological methods, results from these analyses cannot predict how close in time a spoilage community is to the production of clear sensory defects. The main reason for this is that the species composition of a spoilage community does not change dramatically during late shelf life, whereas the ongoing metabolic activities lead to the development of notable sensory deterioration.

ABSTRACT

Microbial food spoilage is a complex phenomenon associated with the succession of the specific spoilage organisms (SSO) over the course of time. We performed a longitudinal metatranscriptomic study on one modified-atmosphere-packaged (MAP) beef product to increase understanding of the longitudinal behavior of a spoilage microbiome during shelf life and beyond. Based on the annotation of the mRNA reads, we recognized three stages related to the active microbiome that were descriptive of the sensory quality of the beef: acceptable product (AP), early spoilage (ES), and late spoilage (LS). Both the 16S RNA taxonomic assignments from the total RNA and functional annotations of the active genes showed that these stages were significantly different from each other. However, the functional gene annotations showed more pronounced differences than the taxonomy assignments. Psychrotrophic lactic acid bacteria (LAB) formed the core of the SSO, according to the transcribed reads. Leuconostoc species were the most abundant active LAB throughout the study period, whereas the transcription activity of Streptococcaceae (mainly Lactococcus) increased after the product had spoiled. In the beginning of the experiment, the community managed environmental stress by cold-shock responses, which were followed by expression of the genes involved in managing oxidative stress. Glycolysis, the pentose phosphate pathway, and pyruvate metabolism were active throughout the study at a relatively stable level. However, the proportional transcription activities of the enzymes in these pathways changed over time.

IMPORTANCE It is generally known which organisms are the typical SSO in foods, whereas the actively transcribed genes and pathways during microbial succession are poorly understood. This knowledge is important, since better approaches to food quality evaluation and shelf life determination are needed. Therefore, we conducted this study to find longitudinal markers that are connected to quality deterioration in a MAP beef product. This kind of RNA marker could be used to develop novel types of rapid quality analysis tools in the future. New tools are needed, since even though SSO can be detected and their concentrations determined using the current microbiological methods, results from these analyses cannot predict how close in time a spoilage community is to the production of clear sensory defects. The main reason for this is that the species composition of a spoilage community does not change dramatically during late shelf life, whereas the ongoing metabolic activities lead to the development of notable sensory deterioration.

INTRODUCTION

There has been an increasing market trend for case-ready meat, i.e., meat that is processed, packaged, and labeled at a central meat-processing facility and delivered to the retail store ready to be put directly into the meat case. From the microbial ecology perspective, these packages are man-made niches in which microbial succession leading to spoilage occurs under the selective pressures of cold and the modified atmosphere (MA) applied (1). In comparison to the atmosphere, increased CO₂ levels (>20%) are used to limit the growth of aerobic food spoilage organisms, whereas high O₂ concentrations (70% to 80%) are needed to keep myoglobin oxygenated to ensure the appealing red color of beef.

Even though case-ready meat products have great market value, we know relatively little about which pathways are active in a developing spoilage community over the course of time during shelf life. The use of CO₂-containing atmospheres combined with refrigeration results in the dominance of psychrotrophic lactic acid bacteria (LAB), Brochotrix thermosphacta, and, to some extent, Enterobacteriaceae in modified-atmosphere-packaged (MAP) meat (1, 2). During the last few years, 16S rRNA gene amplicon-based approaches have increased our knowledge of community diversity in meat spoilage microbiomes (1, 3–9) or microbiota associated with food production environments (4, 7). However, no attention has yet been given to the transcriptional activities related to the succession of a food spoilage community during shelf life.

We recently studied system level responses of three specific spoilage organisms (SSO), i.e., Leuconostoc gelidum subsp. gasicomitatum, Lactococcus piscium, and Paucilactobacillus oligofermentans, during growth in vitro (10) by comparing their time course transcriptome profiles obtained during growth. The study revealed how these LAB employed different strategies to cope with the consequences of interspecies competition. The fastest-growing bacterium, L. gelidum subsp. gasicomitatum, enhanced its nutrient-scavenging and growth capabilities in the presence of other LAB through upregulation of carbohydrate-catabolic pathways, pyruvate fermentation enzymes, and ribosomal proteins, whereas the slower-growing L. piscium and P. oligofermentans downregulated these functions (10). The current study was prompted by the observations we made of the different behaviors of the three LAB while they were growing as communities in vitro (10). We wanted to investigate whether we could recognize similar phenomena in the behavior of the SSO growing on meat over the course of time.

To create a comprehensive view of the activities of a spoilage community over the course of time, we analyzed the metatranscriptomes of natural beef spoilage communities developing at 6°C. We wanted to show which metabolic pathways and defense responses are transcribed during microbial succession and spoilage stages in this commercial meat product. Our aim was to study whether we could distinguish different stages related to sensory deterioration using longitudinal metatranscriptomic analysis.

RESULTS

We followed the development of a microbial community for 11 days using packages originating from the same production lot of a commercial MAP beef product manufactured by a large-scale operator. “Production lot” refers to 1 day’s production and consisted of over 1,000 kg of meat. The experimental time span covered the shelf life and 2 days beyond. Altogether, a total of 65.8 Gb of sequence data (413.7 million reads) was obtained from the samples (see Table S1 in the supplemental material). Good-quality RNA suitable for metatranscriptomic analyses was obtained from all samples from day 2 on. Alongside the total-RNA and rRNA-depleted RNA sequencing, the approaches included conventional microbiology, sensory, and pH analyses associated with meat quality.

Bacterial levels and sensory analysis results.

On the first sampling day, bacterial levels were typical for a fresh product (log 4.6 and log 4.5 CFU/g for LAB and total aerobic count, respectively), and at the end of the experiment, the concentrations had increased to log 8.3 and log 8.2 CFU/g for LAB and total aerobic bacteria, respectively (Fig. 1). During the 11 days, the pH of the meat dropped slightly, from 5.9 to 5.5 (see Table S2 in the supplemental material), and microbial metabolism had also changed the packaging gas composition: O₂ content decreased from 74% to 57%, and CO₂ increased from 20.5% to 37%. These changes were indicative of the activity of psychrotrophic heterofermentative spoilage LAB, as anticipated.

FIG 1.

Bacterial concentration (PCA, total bacterial counts; MRS, LAB counts) and main sensory findings during the experiment. The three stages related to the product were AP (days 3 to 5), ES (days 6 to 9), and LS (days 10 and 11). The curve presents mean values obtained from analyses of two packages, and the error bars represent ranges.

For the bioinformatic analyses, the samples were divided into three groups based on the sensory analysis results (Fig. 1; see Table S2) to enable the detection of possible biomarkers associated with the commercial meat quality of the product during its shelf life and beyond. The panelists started to deem the meat spoiled from day 5 on, with both odor and appearance receiving mean scores of 3/5. Samples assigned to the early-spoilage (ES) group (n = 8 packages) received mean scores of 2.7/5 and 3.0/5 for odor and appearance, respectively. Samples receiving scores below 2 were assigned to the late-spoilage (LS) group, and the mean odor and appearance scores were 1.75/5 and 1.65/5, respectively (Fig. 1; see Table S2).

Active-microbial-community composition during shelf life and beyond.

Taxonomic assignment of the 16S rRNA gene fragments picked from the total RNA fraction indicated that the active microbial communities were composed mainly of Firmicutes from the families Leuconostocaceae and Streptococcaceae and unclassified bacilli (Fig. 2). During the first 5 days of the study (the acceptable-product [AP] stage), Bacillales (unclassified bacilli and Bacillales) were active, showing an abundance of 7.6% of the 16S rRNA gene transcripts, but the abundance decreased beyond day 6, in ES. Leuconostocs were active throughout the experiment (Fig. 2), with an abundance of 16.6 to 17.4%. In ES, the abundance of RNA transcripts of Streptococcaceae, mainly comprising sequences from the genus Lactococcus, increased to 5.2% of all detected 16S rRNA gene transcripts, and their abundance remained high until the end of experiment, i.e., 14.4% in LS. Based on the 16S rRNA results for the abundant genera and previous literature, we selected three SSO, i.e., L. gelidum subsp. gasicomitatum (LMG 18811^T), L. piscium (MKFS47), and Dellaglioa algida (DSM 15638^T), for mapping of all reads against their genomes. The abundance of these psychrotrophic LAB was supported by Bowtie2 mapping, since the majority of all RNA reads were assigned to these species (Table 1). The proportion of L. gelidum subsp. gasicomitatum reads was high from day 3 on (Table 1), whereas L. piscium reads increased on day 7, when the products had been deemed spoiled by the sensory panel. The reads mapping to D. algida showed the highest abundance on day 4 (7.0 and 19.9%) but decreased during storage, and thus, the species was not very active during late shelf life and beyond (Table 1).

FIG 2.

(A) Assignment of 16S rRNA reads to taxa using the Silva 138 database. The results for the most abundant families are shown (the y-axis scale represents relative abundance). (B) Change in activity during the experiment based on read annotation against SEED subsystems (the y-axis scale represents relative abundance). (C and D) Clustering of samples from different time points in an NMDS plot with 16S rRNA gene assignment (C) and by SEED database annotations (D). Samples from day 1 were discarded from the analysis due to the low number of reads recovered from sequencing. The samples were clustered into AP, ES, and LS. The blue lines show the labeled group centroids of the samples, and the circles around the centroids are the 80% confidence areas for standard deviation of the centroids.

TABLE 1.

Percentages of trimmed sequence reads from the mRNA fraction mapping to the genomes of three selected specific spoilage bacteria^a

Sample	% of reads mapping to genome of:			Sum (%)
	L. gelidum subsp. gasicomitatum	L. piscium	D. algida
Day 2-1	3.2	0.1	1.1	4.4
Day 2-2	2.5	0.1	0.5	3.1
Day 3-1	19.7	0.7	7.1	27.5
Day 3-2	30.0	0.8	5.9	36.7
Day 4-1	27.7	3.9	7.0	38.6
Day 4-2	40.7	1.3	19.9	61.9
Day 5-1	43.7	1.7	6.4	51.8
Day 5-2	49.4	6.1	7.8	63,3
Day 6-1	39.7	5.9	4.8	50.4
Day 6-2	27.9	6.5	3.1	37.5
Day 7-1	20.5	17.4	4.2	42.1
Day 7-2	25.4	10.9	2.8	39.1
Day 8-1	37.8	10.4	5.4	53.6
Day 8-2	51.0	7.9	5.1	64.0
Day 9-1	15.5	8.7	4.0	28.2
Day 9-2	32.7	7.9	4.5	45.1
Day 10-1	13.2	2.4	1.7	17.3
Day 10-2	15.9	7.0	2.4	25.3
Day 11-1	18.6	4.0	1.4	24.0
Day 11-2	8.0	4.8	1.7	14.5

^aL. gelidum subsp. gasicomitatum (LMG18811^T), L. piscium (MKFS37), and D. algida (DSM 15638^T). Two packages were opened for each point of analysis.

The three stages from the viewpoints of taxonomy and functional-gene annotation.

To explore the functional genes expressed over the course of time during shelf life, the sequenced mRNA fraction was analyzed. The results from read annotation against the SEED subsystems database (see below) showed that the community transcription activity changed over time in the sample clusters related to stages AP, ES, and LS. In a nonmetric multidimensional scaling (NMDS) plot of the Bray-Curtis distance matrix, the samples clustered according to the stage. A permutation-based multivariate analysis of variance (PERMANOVA) used to analyze both the 16S rRNA gene taxonomic assignments and functional-gene annotations revealed a significant (R² = 0.33; P < 0.001) effect of the stage (AP, ES, or LS) on the taxonomic grouping of the samples (Fig. 2). However, the differences were more pronounced for the functional-gene annotations (R² = 0.67; P < 0.01) (Fig. 2) than for taxonomic assignments.

Specific transcription activities and changes in acceptable product.

The microbial community at the AP stage expressed cold-shock genes (Fig. 3), indicating temperature stress under 6°C storage conditions. Genes related to protein biosynthesis constituted from 16% to 22% of all reads during exponential growth while decreasing to 12% and 9% in the later stages (see Table S3 in the supplemental material). The active growth and succession of the community was thus strongly related to the stage while the product was still acceptable. According to SEED subsystem groups, cell division (1.1 to 1.25%) (see Table S3) was another function expressed significantly highly during the exponential stage. These findings agree with the cultivation results, indicating active community growth in the AP stage.

FIG 3.

Relative abundances (y axis) of transcripts involved in different SEED subsystems. Shown are cold-shock (A), oxidative-stress (B), and respiration (C) categories in AP, ES, and LS with statistically different abundances. Note that each plot has a different y-axis scale. The median values are shown as lines within the boxes, and the mean values are shown as stars. The boxes indicate the 25th and 75th percentiles of the data, and the whiskers extend to the most extreme values with 1.5× interquartile data.

During the AP stage, the abundance of gene transcripts clustering in “carbohydrate metabolism” was one of the highest throughout the study and increased over time (Fig. 2). Detailed comparison of the transcribed reads related to carbohydrate metabolism in the three community growth stages showed that, despite the rather similar expression levels of the fermentation pathways, the relative abundances of genes transcribed in the pathways varied over time (Fig. 4, glycolysis and pentose phosphate pathways). At the AP stage, the bacterial community required energy through fermentation via the pentose phosphate pathway (Fig. 4). Pyruvate metabolism was the second most abundant carbohydrate metabolism pathway during the AP stage. However, unlike the spoilage stages, the majority of the transcripts were assigned as acetate kinase present in several different metabolic pathways. In the spoilage stages, the relative abundance of formate C-acetyltransferase, converting pyruvate to acetyl-coenzyme A (CoA) and formate, was significantly higher (Fig. 4). Additionally, the third most abundant KEGG Orthology (KO) pathway was glycolysis, which had highly expressed genes throughout out the experiment (Fig. 4).

FIG 4.

Carbohydrate metabolism activities of selected main pathways. The relative abundances of genes in pathways involved in carbohydrate metabolism in AP, ES, and LS stages are denoted by the circles, with the circle size relative to the abundance of transcripts. The pie charts visualize the abundances of expressed genes in each stage and the differences in gene transcript abundance. The most abundant genes of each pathway are shown and are organized in clockwise order.

Mapping of the transcripts to the three selected SSO genomes showed that the cold-shock and universal-stress pathways were active during the AP stage (see Table S4 in the supplemental material). In L. gelidum subsp. gasicomitatum, the pentose phosphate pathway was expressed. In addition, amino acid synthesis was abundantly expressed. However, in L. piscium, genes related to lactate fermentation and glycolysis were expressed in relation to the requirement for energy (see Table S4).

Specific transcription activities and changes in samples showing early spoilage.

After day 6, during the ES stage, the cold-shock genes were no longer expressed (Fig. 3), and the succession had apparently led toward the cold-tolerant species and strains. The 16S rRNA data show an increase in the proportion of psychrotrophic LAB with a decrease in mesophilic species. On the functional side, there was a change, as well: the abundance of transcripts involved in reaction to oxidative stress increased, together with genes involved in microbial respiration (Fig. 3).

The members of the ES microbial community expressed genes involved in respiration, as the activities of different cytochrome oxidases and ubiquinone menaquinone-cytochrome c reductase complexes (see Table S3) acting in electron-donating and accepting reactions were found to be abundant. The pathways with higher abundance in expression than those in the AP stage included formation of sugar alcohols, fermentation, and central carbohydrate metabolism (see Table S3). In more detail, the microbial communities expressed genes for glycolysis and gluconeogenesis and the pentose phosphate pathway (Fig. 4; see Table S3), as well as different fermentative energy requirement reactions (see Table S3). Although stress response as a whole increased only slightly over time, the genes involved in oxidative stress were found to be more actively expressed from day 6 onward (Fig. 3). Similar to cold shock reactions, the abundance of gene transcripts involved in cell division and the cell cycle, motility and chemotaxis, nucleoside and nucleotide utilization, and protein metabolism decreased with time (see Table S3). This indicated that the community was no longer growing actively.

In the ES stage, L. gelidum subsp. gasicomitatum was abundant (48%, based on Leuconostoc 16S rRNA gene abundance [Fig. 1]), with 35% of the reads mapping to the genome of L. gelidum subsp. gasicomitatum [Table 1]), and based on mapping of the RNA transcripts to the genome, the active pathways included the pentose phosphate pathway for energy requirements. Transcription of the genes associated with stress and thioredoxin reductase increased compared to the AP stage. Thioredoxin reductase has been shown to be related to survival under oxygen-induced stress conditions (11), which indicates that the species were actively using mechanisms to tolerate the stress caused by the high-oxygen MAP. In addition, heme-dependent respiration of L. gelidum subsp. gasicomitatum (cydA and cydB genes) was found to be active in the AP and ES stages (see Table S3).

At the ES stage, the second most abundant LAB species, L. piscium, utilized NoxE NADH oxidases, which can indicate a way to survive redox stress in the high-oxygen packaging during the active growth phase. Additionally, the aldehyde-alcohol dehydrogenases (AdhE), reported to possess moonlight functions, such as oxidative-stress protection (12), were highly expressed (0.31% ± 0.21%). Unlike the heterofermentative L. gelidum subsp. gasicomitatum, which uses a phosphoketolase pathway for energy requirements, homofermentative L. piscium utilized glycolysis, together with lactate and pyruvate fermentation (see Table S4). Additionally, for L. piscium in ES, the cold-shock genes were highly active, as well as the genes corresponding to sugar transporters. Moreover, the L. piscium gene yngB, coding for the fibronectin-binding protein, was expressed at the ES stage (see Table S4).

Specific activities and changes associated with late spoilage.

In pentose phosphate metabolism, the fermentation of xylulose to pentose (xfp and xpk; xylulose-5-phosphate/fructose-6-phosphate phosphoketolase) increased in the spoiled product, a phenomenon that is associated with the production of pyruvate (see Table S3). In addition, there was increased transcription activity of the pathway leading to the production of ethanol and acetate instead of the common spoilage volatiles diacetyl and 2,3-butanediol. Similarly, the expression of the pentose metabolism gene pflD (formate C-acetyltransferase) increased in the LS stage. The gene catalyzes the reversible conversion of pyruvate into formate and is thus involved in the supply of pyruvate to the pentose phosphate pathway/metabolism.

The ES and LS stages differed in the utilization of di-and oligosaccharides and trehalose, which were both significantly higher in samples from after the use-by date. Trehalose acts as a stress protectant and storage carbohydrate, and it is known that rapidly growing cells having lower trehalose levels than slow-growing and stationary-phase cells (13), explaining the lower transcription activity in exponential-phase cells. Malate metabolism increased significantly in the LS stage, especially the malate dehydrogenase, citrate synthase, and isocitrate dehydrogenase genes associated with the formation of citrate. During the LS stage, starch metabolism was at its highest expression level (Fig. 4, circles), and also, the relative abundance of genes transcribed in relation to it varied greatly over time (Fig. 4, pies).

In addition to the subsystems group “carbohydrate metabolism,” “amino acid metabolism” (arginine, urea cycle, and polyamines) increased with time, as we found a 7-fold increase from ES to LS after the end of the shelf life stage (see Table S3). The genes include arcA, arcB, arcC, arcT, and arcD, corresponding to arginine deiminase, ornithine carbamoyltransferase, carbamate kinase, aspartate aminotransferase, and arginine/ornithine antiporter.

After the product had spoiled at the LS stage, the majority of the reads mapping to L. gelidum subsp. gasicomitatum were involved in a pentose phosphate pathway producing pyruvate, which in the presence of oxygen has been reported to lead to production of diacetyl (9, 14), a spoilage metabolite, through the chemical reaction of alpha acetolactate. As in ES, genes involved in general and oxygen-induced stress were active. However, unlike in ES, proteolytic genes (clpL, encoding ATP-dependent Clp protease) became active, and the abundance of reads mapping to sugar transporters increased. As was seen at the microbial community level, trehalose metabolism increased at this stage, and the genes were assigned to the main bacterium of the stage, L. gelidum subsp. gasicomitatum. For L. piscium, the metabolism did not change between the ES and LS stages.

Based on the MetaPhlAn2 annotation (15), Ascomycota, especially yeasts from the genus Debaryomycetaceae, increased during shelf life (Fig. 5). The proportion of reads mapping to Debaryomycetaceae was 1.49% (±2.37%) in the AP stage and 6.11% (±4.38%) in ES and increased to 25.62% (±12.74%) in LS. Based on the MG-RAST (see below) annotations, the reads from LS mapping to Debaryomycetaceae fungi had active energy metabolism/carbohydrate metabolism, as genes involved in glycolysis, the glyoxylate cycle, and acetate metabolism were among those expressed (see Table S5 in the supplemental material). Besides carbohydrate metabolism, fatty acid metabolism was active, as the genes encoding glycerol-3-phosphate dehydrogenase, stearoyl-CoA desaturase, and acyl-CoA synthetase 2, to mention a few, had RNA transcripts mapping to them (see Table S5). The activity of these yeasts was supported by finding RNA transcripts for several translation elongation factors, as well as 60S and 40S subunit ribosomal proteins. In yeasts, the genes involved in respiration were active, too, and included different cytochrome c oxidases and oxidoreductases, as well as ferredoxin-dependent glutamate synthase. In addition, transaldolase, a gene from the nonoxidative phase of the pentose phosphate pathway, was active in yeasts during LS.

FIG 5.

Proportions of reads belonging to the ascomycotal family Debaryomycetaceae based on Humann2 annotation. Error bars represent standard deviations.

DISCUSSION

Based on the clustering of active transcripts in NMDS and the observed sensory characteristics and bacterial growth levels, the active microbiome and active genes in this MAP beef product had three stages that were descriptive of the quality of the product: AP, ES, and LS. All these stages shared species and functions, but there were several distinct metabolic and other responses associated with each of the three stages (Fig. 2 and 6). As is typical for a packaged meat product (16), the microbial levels of the four stages did not directly correlate with the sensory quality of the product, whereas each of the three stages had specific transcription activities based on the gene level analyses. Starting from the end of the AP stage, the same bacterial species remained active, something that highlights the reason why the active genes and pathways should be analyzed further instead of the active microbes to understand which metabolic pathways play a major role in food spoilage.

FIG 6.

Most active stress responses and phyla in three stages, i.e., acceptable product, early spoilage, and late spoilage, of a meat spoilage community developing in a high-oxygen modified-atmosphere-packaged beef product. The vertical position corresponds to the relative abundance of transcriptional activity.

Even though leuconostocs, lactobacilli, and lactococci have been associated with meat spoilage in several different studies using cultivation- and DNA-based approaches (17–20), little has been known about the most active functions during and after their succession. Based on analysis of the active 16S rRNA genes in the total sequenced RNA fraction, leuconostocs were the most abundant bacteria throughout the study. However, the proportion of psychrotrophic Lactococcus increased after day 6. After shelf life (day 9), the LAB community was still active, but the abundance of RNA transcripts of yeasts from the genus Debaryomycetaceae increased. Yeasts have not been considered to play a role in the spoilage of cold-stored meat before. Some cultivation-based studies have shown the fungi from Saccharomycetaceae phyla present in different meat environments, and also during the ripening of salami (21). The Debaryomycetaceae yeasts increased drastically in the LS stage, while the transcription activity of the spoilage microbial community pertained to alcohol production from the pentose phosphate pathway. The yeasts were obviously active members of the fermentative community, with up to 43.2% relative abundance in total microbial reads on the last day of the experiment (day 11), even though there was variation between different packages. Our results thus show that during a long shelf life, fungi and their spoilage functions increased in this MAP product.

Fermentation of carbohydrates was the most abundant metabolic activity subsystem group detected in the samples over time, as the prevalence of transcribed genes involved in carbohydrate metabolism was high throughout the study. The prevalence of actively expressed genes involved in carbohydrate metabolism showed that, unlike in the spoilage of nonpackaged meat caused by pseudomonads under atmosphere (22), the exhaustion of glucose is not likely a limiting factor for spoilage microbiota under an MA containing 20% carbon dioxide and 80% oxygen. We observed activities of homofermentative LAB, catabolizing glucose using the Embden-Meyerhof-Parnas (EMP) pathway, together with heterofermentative LAB, using the pentose phosphate or pentose phosphoketolase pathway. During the AP stage, when the community was in the exponential growth phase, pyruvate metabolism, especially formate C-acetyltransferase, was expressed. In LAB capable of homofermentative metabolism, pyruvate formate lyase (PFL) is used during the transformation to mixed acid formation under glucose and oxygen limitation to increase the ATP yields (23). Here, the beef was packaged under high-oxygen MA, and the enzyme is oxygen sensitive, because it is cleaved and inhibited in the presence of oxygen (24). Nevertheless, the bacteria were found to express the pfl gene.

The high-oxygen MA and storage at cold (6°C) temperatures created stress for the bacterial communities (Fig. 3). The cold-shock genes were found to be active during the first days of shelf life, before the communities reached the exponential growth phase. However, in the case of L. piscium, cold-shock genes were still expressed during ES, but the growth of this species was delayed in comparison to that of L. gelidum subsp. gasicomitatum. In the lag phase, and especially after the commercially determined shelf life was over, oxygen-induced stress responses increased, and thus, the microbial community members had to strive to cope with the high-oxygen MA. Heme-dependent respiration is among the ways to cope with oxygen-related stress, and also, the relative abundance of transcripts involved in respiration increased. Based on previous functional-genomics studies with single species (10, 14, 19), we hypothesized that in addition to the rapid onset of fermentation, the heme-dependent respiration of leuconostocs may also play a major role in fitness under a high-oxygen MA at the community level. Respiration has been shown to result in increased biomass of LAB and several changes in metabolism and long-term survival (14, 25, 26). The increase in respiration with time indicates that oxygen stress becomes more evident for the community. Interestingly, respiration genes were transcribed in both bacterial and fungal members of the community in LS. Since the transcription of genes related to respiration increased over time and was highest in the AP and ES stages, the activity can be considered to also play a role in long-term survival and viability in a meat spoilage community during shelf life in our study. In addition to the different cytochrome oxidases found in LAB, ubiquinone menaquinone-cytochrome c reductase complexes acting in electron-donating and -accepting reactions were also active in the ES microbial community. In our recent coculture study (10), we noticed a similar trend in respect to L. gelidum subsp. gasicomitatum. It was the fastest-growing bacterium in the coculture containing L. piscium and P. oligofermentans (10). L. gelidum subsp. gasicomitatum enhanced its nutrient-scavenging and growth capabilities in the presence of other LAB through upregulation of carbohydrate-catabolic pathways, pyruvate fermentation enzymes, and ribosomal proteins (10). These findings are in line with those of the present study, highlighting the active role of this SSO in the community over the course of time. D. algida, a psychrotrophic LAB isolated from vacuum-packaged beef, has been isolated from various beef products (27) but was not abundant in the beef product studied (Table 1). The 80% O₂ clearly prevented the growth of the bacterium. Also, in our previous study, where both high-oxygen MAP and vacuum packaging were used, the growth of D. algida was higher in the latter (9). Thus, high-oxygen MAP is likely to support the growth and activity of psychrotrophic LAB from the genera Leuconostoc and Lactococcus.

At the ES and LS stages, an increase in the expression of the noxE gene was observed. As the microbes grow actively while reaching the stationary phase, the use of carbohydrates leads to a lack of potential electron acceptors. The noxE gene has been found to act in lowering the redox potential and thus enabling active metabolism (28). Based on the mapping, the gene was found to be active in L. piscium, and thus, the species can lower the redox potential of its environment to create more optimal growth conditions in MAP beef. The aldehyde-alcohol dehydrogenases were expressed at the ES and LS stages. In Escherichia coli, this gene (adhE) has been shown to have antioxidant activity (29). Also, the adhE genes have been reported to possess moonlight functions, such as oxidative-stress protection (12, 29).

The gene yngB of L. piscium, encoding fibronectin-binding protein, was actively transcribed in ES. The gene has been associated with heat stress in Lactococcus lactis, but here, we found indications of increased expression under oxidative stress on beef. In L. piscium, yngB (locus tag LACPI_1373 [30]) encodes a fibronectin- and collagen binding protein that belongs to a class of fibronectin-binding proteins without a signal peptide. Indications of the gene being used in both fibronectin/collagen binding and biofilm formation have been documented for a Streptococcus suis ortholog (31) and Bacillus subtilis (32).

In ES, sugar alcohol metabolism increased. The substrate for metabolism may originate from the bacterial cell membranes, since Leuconostoc and other LAB have genes for glycerophospholipid metabolism. Another possibility is that polyols produced by the yeast community increased sugar alcohol metabolism (33).

In addition to carbohydrate metabolism and stress-induced reactions, we were able to capture novel information on amino acid metabolism during meat spoilage. The arginine deimination is of interest, since in Lactobacillus sakei, arginine deimination has been shown to increase tolerance of both low pH and acid stress (34). Agmatine can be produced from arginine by a broad range of other bacteria through decarboxylation. The agmatine deiminase pathway, present in, e.g., L. piscium (30), can help bacteria to protect against environmental stresses, including low pH, similarly to how the arginine deiminase pathway does (35).

Conclusions.

Since sensory changes do not follow bacterial concentrations linearly, more accurate approaches for detecting food spoilage are needed. Further studies are needed to determine if the metabolic and other transcription activity markers we detected prove useful in the evaluation of product freshness. A shift toward a Lactococcus-dominated community might also be an indication of the end of shelf life for some meat products, as well as the appearance of yeasts. In addition, the change in stress responses is worth future study, since the expression of genes involved in cold-shock reactions decreased closer to the end of shelf life, whereas the activity of genes involved in respiration and oxygen stress became more pronounced when the product had spoiled. Unlike what was previously thought, carbohydrate metabolism was one of the most active functions throughout time and in the change from fresh to spoiled product. This might indicate that carbohydrate concentrations do not limit the growth of meat spoilage communities as much as previously thought.

MATERIALS AND METHODS

Experiment description.

Twenty-two packages of tenderized beef loin fillets stored in MAP were bought in commercial packaging. The samples were collected from the same lot (comprising 1 day’s production) from a large-scale producer. During the experiment, the samples were stored at 6°C. Two packages were sampled destructively each day for 11 days. Before opening, the concentrations of O₂ and CO₂ in the package gas were measured (Checkpoint; PBI Dansensor) through a double septum and, after opening, one fillet (approximately 120 g) from each package was immediately transferred to a stomacher bag with 10 ml of RNAlater (Ambion) and 5 ml of peptone water. The bacterial cells were removed from the beef surface with a stomacher (Stomacher 400; Seward, West Sussex, United Kingdom) at low power for 30 s. The liquid was collected from the stomacher bag, and cell collection was conducted via two-stage centrifugation to separate the mammalian and bacterial cells. First, the supernatants were centrifuged in 15-ml conical tubes at 200 relative centrifugal force (RCF) for 3 min at 4°C to remove the fat and eukaryotic cells. The supernatant was collected and centrifuged in 1.5-ml Eppendorf tubes at 13,000 rpm for 3 min at 4°C to collect the cells. The supernatant was poured off and replaced with 500 μl of RNAlater. The collected cells were stored at −70°C until nucleic acid extraction was performed.

Another fillet was used for quantitative microbiological and sensory analysis. Serial 10-fold dilution series were made from 22 g of beef with 198 ml of peptone-salt buffer (0.1% peptone, 0.9% NaCl) and homogenized by using a stomacher at medium power for 1 min, followed by cultivation on de Mann-Rogosa-Sharpe (MRS) plates (LAB). In addition, plate count agar (PCA) was used to quantify the total cultivable microbes from the beef fillets. The PCA plates were incubated at 25°C for 3 to 5 days before colony enumeration. The MRS plates were incubated in jars made anaerobic with a commercial atmosphere generation system (AnaeroGen; Oxoid) at 25°C for 5 days, after which the CFU were calculated. Sensory analyses were performed by a trained panel of at least five individuals. For these analyses, the beef samples were equilibrated at room temperature. Beef samples from the same meat lot were stored fresh (day 0) in the freezer (−20°C) and used as a reference. The panelists evaluated the odor and appearance of the samples using a five-point scale (1, severe defect, spoiled; 2, clear defect, spoiled; 3, mild defect, satisfactory; 4, good; 5, excellent); the observed deficiencies were described by the panelists.

RNA extraction and library preparation.

The extraction protocol was modified from that of Chomczynski and Sacchi (36). First, the samples were centrifuged at 16,000 RCF at 4°C for 10 min, and the RNAlater was carefully removed. The pellet was immediately covered with 500 μl of phenol-chloroform-isoamyl alcohol (Sigma) and 500 μl of denaturation solution (4 M guanidium thiocyanate, 24 mM sodium citrate, 0.5% sarcosyl, and 0.1 M beta-mercaptoethanol). The samples were transferred to lysing matrix E tubes (mBio) and beaten with beads in a FastPrep-24 instrument (MP Biomedicals) for 30 s at 5.5 m/s. After the bead beating, the tubes were placed on ice for 5 min, followed by centrifugation at 13,000 RCF at 4°C for 10 min. The supernatant was mixed with 500 μl chloroform, vortexed, and centrifuged at 4°C at 13,000 RCF for 5 min. Nucleic acids were precipitated with 3 M sodium acetate (NaAc) (1:10), a 10× volume of isopropanol, and 1 μl of GlycoBlue (Invitrogen) at −20°C for 2 h. The nucleic acid pellet was washed with 70% ethanol and eluted to 100 μl of diethyl pyrocarbonate (DEPC) water. DNA and RNA were purified with a Qiagen Allprep DNA RNA kit according to the manufacturer’s recommendations with additional DNase treatment. The resulting DNA and RNA quality was checked by gel electrophoresis and with Agilent Bioanalyzer Nano chips, respectively. The concentration was checked with Cubit (Thermo).

The sequencing libraries were made from both the rRNA-depleted libraries (mRNA library) and total RNA. rRNA depletion for the mRNA library was conducted with a Ribozero kit (Epicentre) with the low-input protocol. An Illumina TruSeq stranded mRNA kit was used for library preparation. The samples were sequenced on an Illumina NextSeq at the DNA-sequencing and genomics laboratory at the University of Helsinki.

Quality filtering.

The raw reads were filtered for quality with Cutadapt (37) to remove the reads, with a reverse adapter with quality cutoff Q20 and with a length below 60 bp. The quality-filtered reads were mapped to the Bos taurus genome (accession number NKLS00000000.2) using Bowtie2 (38) with default parameters, and reads mapping to B. taurus were discarded.

rRNA mapping.

The rRNA reads from the quality-filtered non-B. taurus reads of the total RNA libraries were searched with Metaxa2 (39) and subjected to a BLAST search against the Silva 138 database (40) to identify the 16S rRNA reads and thus analyze the change in active species composition of the samples.

Annotation.

All sequence reads from mRNA libraries were annotated against the KEGG (41) and SEED (42) databases with an E value cutoff of 1e−5, a minimum identity cutoff of 60%, and a minimum alignment cutoff of 15 bp at MG-RAST (43). The annotated genes were normalized by sample read number. Three bacterial genomes, i.e., L. gelidum subsp. gasicomitatum (LMG 18811; FN822744.1), L. piscium (MKFS47; NZ_LN774769), and D. algida (DSM 15638; NZ_AZDI01000000), were selected for more thorough analysis. The sequence reads from all the samples were mapped to the UniProt (44) genes of each genome with Bowtie2 (38). In addition, the reads were annotated to species in MetaPhlAn2 (15), and specific pathways were examined in Humann2 (45).

Statistics.

A Bray-Curtis distance matrix of relative-abundance-normalized 16S rRNA picked from the total RNA libraries and metabolic pathway data from the mRNA libraries was ordinated with NMSD in R (46) using Vegan (47), ggplot (48), and Phyloseq (49). Bacterial abundances were matched with the factor “Sample stage” with the function envfit from the Vegan package, and an 80% confidence area was used for standard deviation of the centroids. The statistical differences among the functional profiles of the three time points were analyzed with the STAMP software package (50). An ANOVA with a Games-Howell post hoc test and Storey’s false-discovery rate (FDR) for correction was conducted.

Data availability.

Sequence data were deposited in the European Nucleotide Archive (ENA) under project PRJEB20288. The annotated metagenomes are available at MG-RAST project BEEF_RNA (ID mgp10843).