A Temporal -omic Study of Propionibacterium freudenreichii CIRM-BIA1T Adaptation Strategies in Conditions Mimicking Cheese Ripening in the Cold

Propionibacterium freudenreichii is used as a ripening culture in Swiss cheese manufacture. It grows when cheeses are ripened in a warm room (about 24°C). Cheeses with an acceptable eye formation level are transferred to a cold room (about 4°C), inducing a marked slowdown of propionic fermentation, but P. freudenreichii remains active in the cold. To investigate the P. freudenreichii strategies of adaptation and survival in the cold, we performed the first global gene expression profile for this species. The time-course transcriptomic response of P. freudenreichii CIRM-BIA1T strain was analyzed at five times of incubation, during growth at 30°C then for 9 days at 4°C, under conditions preventing nutrient starvation. Gene expression was also confirmed by RT-qPCR for 28 genes. In addition, proteomic experiments were carried out and the main metabolites were quantified. Microarray analysis revealed that 565 genes (25% of the protein-coding sequences of P. freudenreichii genome) were differentially expressed during transition from 30°C to 4°C (P<0.05 and |fold change|>1). At 4°C, a general slowing down was observed for genes implicated in the cell machinery. On the contrary, P. freudenreichii CIRM-BIA1T strain over-expressed genes involved in lactate, alanine and serine conversion to pyruvate, in gluconeogenesis, and in glycogen synthesis. Interestingly, the expression of different genes involved in the formation of important cheese flavor compounds, remained unchanged at 4°C. This could explain the contribution of P. freudenreichii to cheese ripening even in the cold. In conclusion, P. freudenreichii remains metabolically active at 4°C and induces pathways to maintain its long-term survival.


Introduction
Micro-organisms have a determining contribution to the formation of the typical flavor and texture characteristics of each cheese variety. The metabolism of lactic starters and of adjunct ripening cultures has mainly been studied in conditions mimicking their growth in milk and in cheese. More recently, studies have specifically focused on the metabolism of non-growing bacteria under the sub-optimal conditions prevailing in cheese. They showed that bacterial metabolism is markedly modified by cheese conditions, like starvation and low temperatures. Bacteria expressed genes associated with specific metabolic pathways that could also contribute to the ripening process [1][2][3][4].
Propionibacterium freudenreichii is widely used in Swiss cheese manufacture for its key contribution to the formation of the characteristic holes (or ''eyes'') and flavor during the ripening of this cheese variety [5]. Different combinations of time and temperature are applied during the ripening of Swiss cheeses, according to the manufacturers' practices. Typically, cheeses are ripened for 2-3 weeks at 10-14uC, then in a warm room (20-24uC) until the cheeses have enough holes, and are finally transferred to a cold room (4-6uC). P. freudenreichii grows during the ripening in the warm room with populations reaching stable levels over 10 9 colony-forming units (cfu)/g [1]. It converts the lactate produced by lactic acid bacteria into propionate, acetate and CO 2 . Propionic fermentation mainly occurs in the warm room and is markedly slowed down in the cold room. However, P. freudenreichii is known to remain active during the storage of cheese at 4uC, as recently shown by real time reverse transcription PCR [1]. In particular, it goes on producing some flavor compounds in the cold, like short branched-chain acids from branched-chain amino acids catabolism, and free fatty acids from milk fat hydrolysis [6]. The aim of the present study was to better understand how P. freudenreichii copes with cold-induced stress and remains active when cheeses are transferred to the cold room. This study was performed under conditions mimicking Swiss cheese ripening, by applying for the first time a whole genome transcriptomic approach to P. freudenreichii. Transcriptomics and other -omic approaches have increasingly been used over the last few years to provide a comprehensive view of microorganism physiology in various research areas. For example, the transcriptomic response of the starter bacterium Lactococcus lactis has recently been studied in conditions mimicking cheese manufacture [7].
The time-course transcriptomic response of P. freudenreichii CIRM-BIA1 T strain was analyzed during its growth at 30uC, then during a further 9 days at 4uC. In addition, temporal changes in the proteome and extracellular metabolome were monitored. Our results show that P. freudenreichii under-expresses many genes involved in cell machinery in the cold, but remains active, redirecting its carbon metabolism towards glycogen accumulation. Interestingly, some genes involved in the formation of important flavor compounds of cheese were not significantly affected in the cold. These results highlight the contribution of P. freudenreichii to the formation of flavor compounds even at low ripening temperatures. Thus, our data provides insights into the strategies used by P. freudenreichii to adapt to a low temperature and to maintain its long-term survival.

Experimental conditions mimicking cheese ripening
In cheeses stored at cold temperatures, P. freudenreichii stops growing, slows down propionic fermentation, but it remains active [1] and continues to produce some metabolites [6]. To better understand how P. freudenreichii CIRM-BIA1 T copes with coldinduced stress when cheeses are transferred to the cold room, anomic approach was applied. The culture conditions were chosen to mimic the medium and the temperature-shift withstood by P. freudenreichii during Swiss cheese ripening.
During Swiss cheese ripening, P. freudenreichii grows in the absence of carbohydrates since all the lactose of milk is converted to lactate by lactic starters within the first day of cheesemaking [5]. Lactate is used as the main carbon source by P. freudenreichii and is available until the end of ripening [5,8,9]. The Swiss cheese aqueous phase is a rich medium, in which the soluble nitrogen matter increases during ripening from about 20 g/kg to 50-150 g/kg, with a proportion of free amino acids reaching up to 50% [9]. To mimic cheese conditions, bacteria were grown in a rich medium (YEL [10]) containing lactate (130 mM) as carbon source and high amounts of soluble nitrogen matter (Table S1), and as follows: at 30uC for 40 h, lactate was then added to prevent its exhaustion, and the cultures were further incubated for 9 days at 4uC (Fig. 1A). Preliminary experiments showed that, in YEL medium, either containing twice as much initial lactate content (260 mM instead of 130 mM), or supplemented with 130 mM lactate during incubation, P. freudenreichii CIRM-BIA1 T continued, or started growing again at the same rate if the temperature was maintained at 30uC, and consumed all the available lactate (Fig.S1). These results showed that low temperature, and not nutrient starvation, was the main factor inducing changes in CIRM-BIA1 T metabolism under our experimental conditions, like in Swiss cheese.
The time-course of growth, lactate consumption and propionic fermentation are shown in Figure 1A. Cells grew at 30uC with a generation time of 5.460.2 h. They consumed lactate and concomitantly produced propionate and acetate, with a ratio of 2.260.1 moles of propionate per mole of acetate, in agreement with literature data [5,11,12]. Pyruvate was excreted during exponential growth when lactate was abundant, and then reused when the lactate concentration was low, at the end of incubation at 30uC (40 h) (Fig. 1B). As previously observed under similar conditions, P. freudenreichii can excrete pyruvate in order to maintain its concentration in the cell under its toxicity threshold [13].
Succinate and NH 3 were also produced, concomitantly with aspartate (Asp) and asparagine (Asn) consumption (Fig. 1B, Table  S1). The molar ratio of succinate, NH 3 , and Asp/Asn was close to the expected ratio of 1. P. freudenreichii is known to deaminate Asp to fumarate which is, in turn, converted to succinate, with a generation of additional energy. Previous studies performed under similar conditions showed that, when extra Asp was added to a complex medium containing lactate, up to 40 mM of Asp were converted to succinate concomitantly with lactate fermentation [11]. In Swiss cheese, Asp and Asn are also metabolized by P. freudenreichii and, consequently, their final content is much lower than in similar cheese varieties without propionic fermentation [8].
In the present study, the concentration of six other free amino acids, glutamate (Glu), glycine (Gly), serine (Ser), arginine (Arg), lysine (Lys) and alanine (Ala) also exhibited a decrease ranging from 0.22 to 1.06 mM during the incubation at 30uC (Table S1). Short chain acids (acetate, propionate and methylbutanoic acids) are considered as flavor compounds in Swiss cheese [6]. Beside acetate and propionate, CIRM-BIA1 T also produced 46.865.2 mM of methylbutanoic acids during its growth at 30uC (Fig. 1C).
All these results showed that the metabolites produced by P. freudenreichii CIRM-BIA1 T during growth in cultures at 30uC were in close agreement with that produced during its growth in Swiss cheese.

Time-course of gene expression over the incubation period
A whole-genome transcriptomic approach was used to investigate the gene expression of P. freudenreichii CIRM-BIA1 T at five sampling times. Overall, 565 genes were detected as differentially expressed (DE genes) with P,0.05 and |(fold change)|.1, at least at one sampling time in comparison with the reference time 20 h. The DE genes represented 25% of the protein-coding genes of CIRM-BIA1 T genome targeted in the microarray. The complete microarray expression data for these genes is presented in Table  S2.
To get information on the time-course of gene expression, the temporal expression profiles of the DE genes were classified using a quadratic regression method, convenient to analyze time-course microarray data [14]. A large majority (80%) of genes exhibited two patterns characterized by i) a marked increase or decrease in expression occurring between the incubation periods at 30uC and at 4uC, and ii) a limited change in expression over the period at 4uC (Fig.S2). In addition, 19% of genes exhibited a linear decrease or increase, and 1% showed a transient up-or down-regulation at 40 h that could not be fitted by this quadratic regression model at P,0.05.
According to these temporal expression patterns, the results will be further discussed by comparing gene expression at two main sampling times, i.e. at the reference time 20 h, corresponding to the exponential phase of growth at 30uC, and after 3 days at 4uC. In agreement, RT-qPCR validation experiments were solely performed at these two sampling times. RT-qPCR was applied to 28 genes observed as differentially expressed in microarray experiments, preferentially chosen (19/28) among the up-regulated genes at 4uC. Overall, the confirmation of microarray data by RT-qPCR showed an excellent agreement between the foldchange values obtained from the two approaches (RT-qPCR fold change = 1.176microarray fold change, R 2 = 0.91), thus validating the results of microarray experiments (Tables 1, 2, 3 and 4). Moreover, the changes in protein expression between 30uC and 4uC were analyzed by using a proteomic approach combining 2D electrophoresis and protein identification by mass spectrometry (Fig. 2, Table S4).

Changes in expression according to gene functional category
Genes were classified in 19 functional categories (Table S3). The number of up-or down-regulated DE genes in each category is illustrated in Figure 3. Only 24 genes (1% of the whole genome) were differentially expressed at the end of incubation at 30uC (40 h), in comparison to the reference time (Fig. 3A), whereas 413 genes were differentially expressed at 4uC (3 days) (Fig. 3B).
At 4uC, most of the DE genes implicated in functions related to the general cell machinery were down-regulated, including those related to the metabolism of nucleotides (Nt), co-enzymes and prosthetic groups (C), energy (E), and proteins (PD, P, PM) (Fig. 3B). In contrast, similar numbers of down and up-regulated genes were observed in some gene categories, like adaptation to atypical conditions (A), DNA metabolism (D), transcription (TS) and transport and metabolism of amino acids (AA), carbohydrates (CH), and inorganic ions (T). Lastly, the category dedicated to phosphate metabolism exhibited exclusively up-regulated genes at 4uC (Fig. 3B). These general trends summarize the strategies of Figure 1. Time-course of P. freudenreichii CIRM-BIA1 T metabolic activity over a 40 h-incubation at 306C followed by a further 9 days at 46C. Lactate was added at 64 h (Q) to mimic cheese ripening conditions. A, growth monitored by optical density (650 nm) measurements (black circle), lactate consumption (cross), production of acetate (white triangle) and propionate (black triangle); B, production of pyruvate (white rhombus) and succinate (black rhombus), consumption of aspartate (white triangle) and asparagine (black triangle); C, production of methylbutanoate (sum of 2-methyl-and 3-methylbutanoate acids   Table S4 and Fig. 2 adaptation and survival of P. freudenreichii CIRM-BIA1 T at 4uC, which are discussed in detail in the following sections.

General cell machinery slowing down
At 4uC, CIRM-BIA1 T growth and general metabolism slowed down. The growth curve flattened out at an OD of 3, i.e. 1.960.5.10 9 colony forming unit/ml (cfu/ml). Accordingly, expression of the ftsX gene involved in cell-division was reduced at 4uC, as revealed by both microarray and RT-qPCR results (fold change of 24.2 and 24.3, respectively) ( Table 1). Lactate consumption and acetate and propionate production rates also decreased by a factor of 19, 29 and 25, respectively (Table 5). However, cells did not show any sign of carbon starvation since cstA, coding for a carbon-starvation protein which was transiently up-regulated at 40 h in response to the low lactate content, was then down-regulated after lactate addition at 4uC ( Table 2).
More than 60% of the DE genes were down-regulated at 4uC. At cold temperatures, many bacteria also exhibit a general slowing down, like for example Lc. lactis in model cheeses when placed at 12uC for seven days [7]. In particular, in the present study, P. freudenreichii CIRM-BIA1 T down-regulated most of the DE genes involved in energetic metabolism (29/34). This included genes encoding ATP synthases (atpABCDEFGH) and several oxidoreductases of the electron transport chain (nuoABCDEFG, nuoJKL, nuoM, cydAB, and sdhAC, Table S2). Genes encoding the elongation factor (tuf) and ribosomal proteins were also downregulated at 4uC, as previously observed in L. lactis [7]. Proteomics also partly confirmed CIRM-BIA1 T metabolism slowing down. For example, NuoD and the elongation factor were some of the identified proteins which volume decreased at 4uC ( Table 1, Table  S4 and Fig. 2). Cells also diminished protein synthesis at 4uC, since 87% (33/38) of the DE genes for this category were downregulated (Table S2).
Concerning amino acid synthesis pathways, all the DE genes were down-regulated at 4uC, except for Arg synthesis (Table S2). For example, genes coding for proteins of the branched-chain amino acid ABC transporter (livFG, brae, brad, ydaO, fold change ranging from 21.3 to 23.3), and for Met, Glu, Trp, Lys Ile, and Val synthesis were down-regulated at 4uC (Table S2).
Concerning carbohydrate metabolism, genes coding for enzymes of the TCA cycle (icd and acn) and components of the pyruvate dehydrogenase (aceE and lpd) were also down-regulated at 4uC. General cold stress response When cultures were placed at 4uC, P. freudenreichii CIRM-BIA1 T had to face a cold stress and significantly up-or down-regulated stress protein-coding genes. Cells up-regulated genes coding for one stress-response transcriptional regulator protein and two cold shock proteins (pspC, cspA and cspB, fold change of +2.6, +2.2 and +2.6, respectively) ( Table 2, Table S2). Cold shock proteins are essential for the cells to resume growth in the cold and are synthesized in many bacteria [15]. These proteins, usually transiently expressed and detected during the acclimation phase, act as chaperones [16] and could constitute an asset for P. freudenreichii to maintain its metabolic activity in the cold. P. freudenreichii CIRM-BIA1 T also down-regulated several chaperone and heat shock protein-coding genes (groSL and dnaKJ operons, hsp20, clpB, grpB), with fold change values ranging from 21.3 to 25.2 over the whole period at 4uC (Table 3, Table S2). The spot volumes of DnaK2, GroS1, ClpB2 and Hsp20 1 proteins also decreased at 4uC (Table 3, Table S4 and Fig. 2). These proteins are known to be induced by several other stresses in P. freudenreichii [17,18], and, in contrast, to be repressed upon cold shock in different bacterial species [19,20], as observed for CIRM-BIA1 T .
The up-regulation of two genes encoding DEAD-box RNA helicases, PFREUD_04260 and PFREUD_13460, with fold change values of +1.0 and +1.9 respectively (Table S2), were other indicators of the cold stress response in P. freudenreichii CIRM-BIA1 T . The up-regulation of these helicases has previously   Table S4. doi:10.1371/journal.pone.0029083.g002 been observed in L. monocytogenes [16] and many other microorganisms in the cold [21]. They help to relieve secondary structures formed by RNA at low temperatures, thus facilitating translation, and could contribute to the ability of P. freudenreichii cells to withstand cheese ripening conditions [21].
The maintenance of membrane fluidity is another absolute priority for bacterial survival in the cold [22]. P. freudenreichii membrane contains a large majority of odd numbered, iso-and anteiso-form branched-chain fatty acids [23,24]. In the cold, P. freudenreichii changes its membrane fatty acid profiles in reducing the presence of iso-form fatty acids in favor of anteiso-form fatty acids [24]. Branched-chain fatty acids are synthesized thanks to the activity of a branched-chain alpha-keto acid dehydrogenase, encoded by the bkd operon. In the present study, the genes of the bkd operon were either non-differentially expressed or were downregulated (bkdA2, fold change = 23.2, Table 2). Similarly, in L. monocytogenes, another bacterial species containing branched-chain fatty acids, the transcription of the bkd operon is not affected at 4uC [16]. This suggests that the major cold regulation point in CIRM-BIA1 T is found downstream of the bkd operon in the branched-chain fatty acid synthesis pathway, as in L. monocytogenes.
In addition, the genes involved in Arg transport and synthesis (argBCDEFG, argJK) were up-regulated at 4uC ( Table 2, Table  S2). This result was unexpected because more than 2 mM of free Arg were still present in the culture medium at the end of incubation (Table S1). Whether the up-regulation of these genes resulted or not in the accumulation of intracellular Arg, and its role for P. freudenreichii remains hypothetic. In many bacteria, Arg can be used as an alternative energy source and to control intracellular pH [25].
Long-term survival mechanism and rerouting of carbon metabolism P. freudenreichii CIRM-BIA1 T kept a high viability over the whole period at 4uC, with 90% of the viable cells enumerated at 3 days being still cultivable at 9 days (1.760.3.10 9 cfu/ml). Cells had to find other ways to get energetic supplies, since genes of the electron transport chain and ATP synthesis were down-regulated at 4uC. P. freudenreichii is known to accumulate inorganic polyphosphates (polyP) as an energy reserve during its growth on lactate [26], which are degraded by inorganic pyrophosphatases [27]. In the present study, an inorganic pyrophosphatase-coding gene, ppa, was especially up-regulated at 4uC (RT-qPCR fold change = +3.3, Table 3). PolyP could thus be used to regenerate ATP from ADP in CIRM-BIA1 T , as previously described in other bacteria [28].
A rerouting of carbon metabolism in P. freudenreichii CIRM-BIA1 T in the cold was also observed, as summarized in Figure 4 and discussed below. The formation of pyruvate from various sources was actively up-regulated. The expression of lactate dehydrogenases, catalyzing the oxidation of lactate into pyruvate, was either maintained at 4uC (ldh1 and dld genes), or significantly up-regulated (ldh2 gene, RT-qPCR fold change = +1.9) ( Table 4). In agreement, lactate was constantly consumed at 4uC, with a total of 55.363.8 mM consumed over the period at 4uC (Fig. 1A). Moreover, the reactions of Ala and Ser conversion into pyruvate were up-regulated. Indeed, the transport of Ala and Ser into the cell, encoded by cycA2, was up-regulated, as well as the conversion of Ala and Ser into pyruvate, encoded by ald and sdaA genes respectively (fold change of +7.3 and +3.5, respectively) ( Table 2,  Table S2, Fig. 4). The Ald protein was also induced at 4uC (Table 2). However, the concentration of free Ala and Ser in culture supernatant decreased by only 0.1560.06 and 0.1860.04 mM, respectively, over the incubation period in the cold (Table S1). Intracellular Ala and Ser stocks may have also been converted into pyruvate. Propionate and acetate continued to be produced with the same propionate/acetate ratio compared to that observed at 30uC (Fig. 1A, Table 5). However, the yield of propionate and acetate produced per mole of lactate consumed tended to decrease at 4uC, compared to 30uC (Table 5). This result suggests that a fraction of the pyruvate produced from lactate was redirected towards other metabolic pathways at 4uC.
Our transcriptomic and proteomic data suggest that P. freudenreichii CIRM-BIA1 T metabolism was redirected towards glycogen synthesis through gluconeogenesis (Fig. 4). Hence, pyruvate phosphate dikinase, catalyzing the conversion of pyruvate into phosphoenolpyruvate (PEP), was over-expressed at 4uC (Table 4). PEP could have been converted to a-D-glucose-6-P via gluconeogenesis, as suggested by the up-regulation at 4uC of eno1, fba2, pgi genes and some of the corresponding proteins (Eno1 and Pgi) (Fig. 4, Table 4). Gluconeogenesis was previously reported in resting cells of the same P. freudenreichii strain incubated Table 5. Indicators of P. freudenreichii CIRM-BIA1 T strain metabolism during a 40 h-incubation at 30uC followed by 9 days at 4uC. in the presence of labeled pyruvate [29]. In our study, the final product of gluconeogenesis, a-D-glucose-6-P, could have been converted into glycogen, as suggested by the up-regulation of glgC (fold change = +2.6) encoding ADP-glucose synthase, the main regulation enzyme of glycogen synthesis [30], and of two other enzymes of glycogen synthesis pathway, at gene and/or protein level (Fig. 4, Table 4). Glycogen is an a-1,4 linked, a-1,6 branched glucose polymer used for long-term energy storage. Its metabolism is highly interconnected with a wide range of cellular processes and tightly adjusted to the nutritional and energetic status of the cell [31]. P. freudenreichii is able to synthesize and accumulate intracellular glycogen, although it is not able to degrade extracellular glycogen [32]. Storing glycogen should then be an asset for P. freudenreichii contributing to its aptitude to adapt, resist and survive to stressful conditions such as the low temperature during cheese ripening.
Aroma compound production in the cold P. freudenreichii has a key role in the formation of Swiss cheese flavor by producing a variety of flavor compounds from three main origins: propionate from lactate fermentation, methylbutanoic acids from the conversion of Ile and Leu, and free fatty acids from milk fat hydrolysis by lipolytic carboxylic ester hydrolases [32].
Most of the genes involved in Wood-Werkman cycle converting pyruvate to propionate were not differentially expressed in conditions mimicking cold room ripening of cheeses. The Wood-Werkman cycle seemed to be maintained at 4uC, as indicated by the increase in spot volume at 4uC of FumC, Cat, MmdA, MutA and MutB proteins ( Table 4, Fig. 4 and Table S4). Accordingly, propionate continued to be produced, as stated above, but its formation rate decreased by a factor of 25 ( Table 5).
The production of methylbutanoic acids, in contrast, decreased by a factor of only 3 at 4uC (Fig. 1C, Table 5). The formation of methylbutanoic acids from Leu and Ile is catalyzed by genes encoding branched-chain amino acid aminotransferases (pabC and ilvE) and genes of the bkd operon [12]. As already mentioned, these genes were not differentially expressed at 4uC, except for the bkdA2 gene.
Ten genes coding for (putative) carboxylic ester hydrolases were previously identified in CIRM-BIA1 T genome [33] and could be involved in fat hydrolysis or ester synthesis. Interestingly, nine of these 10 genes were not differently expressed at 4uC, whereas the remaining one (pf379) was significantly up-regulated (fold change = +1, Table S2). These results coincide with the observation that milk fat hydrolysis and ester synthesis are maintained in cheese at cold temperatures [6].

Conclusions
This study brought new insights into the metabolism of P. freudenreichii CIRM-BIA1 T in conditions mimicking cheese transfer from the warm room to the cold room. It highlights in particular how this strain re-directs some metabolic pathways to accumulate carbon storage compounds and to use alternative energy storage compounds like polyP, contributing to its long-term survival. Further work is in progress to determine if other P. freudenreichii strains exhibit similar metabolic adaptations in the cold, compared to CIRM-BIA1 T , or if they use additional adaptive strategies. Since some strains of P. freudenreichii are now considered for their probiotic properties [34], our results could also be of interest to develop dairy products containing probiotic strains keeping high viable populations during their storage at a low temperature.

Strain and culture conditions
The sequenced P. freudenreichii CIRM-BIA1 T strain [35] was grown in YEL broth [10] under air atmosphere without agitation. Growth was monitored by optical density measurements at 650 nm (OD 650 ). Cultures were incubated at 30uC until they reached an OD 650 of about 260.11 (at 40 h), i.e. before the exhaustion of the main carbon source (lactate). Then, cultures were incubated for a further 9 days at 4uC. Lactate (132 mM/ml, Panreac, Lyon, France) was added after 24 h incubation at 4uC in order to mimic the lactate availability during cheese ripening, thus preventing carbon starvation. Cultures were harvested at five stages: in the middle of the exponential growth phase at 30uC (20 h, OD 650 <0.5), at the end of exponential growth phase at 30uC (40 h, OD 650 <2), and after 3, 6 and 9 days of incubation at 4uC. After 3 and 9 days at 4uC, the bacterial concentrations were determined by CFU counting on YEL agar incubated at 30uC for 5 days, in anaerobic conditions. Experiments were performed in triplicate as independent cultures.

RNA extraction
Total RNA extraction was adapted from Deutsch et al [36]. Briefly, at each sampling time, cells were harvested (80006g, 10 min, room temperature or 4uC regarding the sampling time). Cell pellets were suspended in 100 ml of TE buffer 16 (Qiagen, Courtaboeuf, France) containing lysozyme (20 mg/ml), and incubated for 15 min at 24 uC before addition of 350 ml RLT buffer (from the Qiagen RNeasy minikit). 50 mg of zirconium beads and 100 ml of sodium dodecyl sulfate (SDS, 20%) were then added and cells were disrupted for 2690 seconds at 30 Hz by using a Retsch MM301 high-speed mixer mill (Grosseron, France). RNA was extracted according to the instructions of the RNeasy kit. Then DNase treatments of the RNA samples were performed using a DNA-free kit (Ambion, Texas, USA) according to the manufacturer's instructions. RNA quantification and contamination by proteins were assessed using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Inc., Rockland, DE). RNA quality was evaluated using an Agilent 2100 bioanalyzer (Agilent Technologies, Santa Clara, CA) and the RIN value was calculated as previously described [37]. All of the RNA samples had a RIN value greater than 8.4, indicating a good RNA integrity. The absence of genomic DNA was confirmed by quantitative PCR (qPCR).

Transcriptomic analyses and statistical analysis
Microarray design, manufacture, synthesis and image scanning were provided by Source BioScience imaGenes (Berlin, Germany). Briefly, P. freudenreichii CIRM-BIA1 T microarrays were designed as 8615K (i.e. 8 samples per slide with 15000 oligos targeting each sample) custom microarrays including all ORFs and intergenic regions from the whole genome of P. freudenreichii CIRM-BIA1 T strain. So, the number of oligos targeting CIRM-BIA1 T strain genome ranged from 1 to 5, with a mean of 4 oligos per gene or intergenic region. The complete microarray design is available in NCBIs GEO platform under accession number GPL13959. At each sampling time, hybridizations consisted in three biological replicates. A technical replicate was also performed for the reference time at 20 h. Cyanine-3 (Cy3) labeled cRNA was prepared from (50-100 ng) RNA using the Low Input Quick Amp Labeling Kit (Agilent Technologies) according to the manufacturer's instructions, followed by RNAeasy column purification (Qiagen). Dye incorporation and cRNA yield were checked with the NanoDrop ND-1000 Spectrophotometer. 600 ng of Cy3labelled cRNA (specific activity .10.0 pmol Cy3/ug cRNA) was fragmented (60uC, 30 min) in a reaction volume of 25 ml containing 16 Agilent fragmentation buffer and 26 Agilent blocking agent following the manufacturer's instructions. On completion of the fragmentation reaction, 25 ml of 26 Hi-RPM hybridization buffer (Agilent Technologies) was added to the fragmentation mixture and hybridized to Agilent specific 8615 K microarray for 17 h at 65uC in a rotating Agilent hybridization oven. After hybridization, microarrays were washed for 1 min at room temperature with GE Wash Buffer 1 (Agilent Technologies), then for 1 min with 37uC GE Wash buffer 2 (Agilent Technologies), and finally for 1 min at room temperature with acetonitrile. Slides were scanned immediately after washing on the Agilent DNA Microarray Scanner (G2565CA) using one color scan setting for 8615 K array slides (Scan Area 61621.6 mm, Scan resolution 5 mM, Dye channel set to Green and Green PMT set to 100%). The scanned images were analyzed with Feature Extraction Software 9.1 (Agilent Technologies) using default parameters. Normalization and microarray analysis were performed using limma package [38] of the R software (www.Rproject.org). All intensities were expressed as log 2 of the normalized signal intensities. Differential analysis was performed using empirical Bayes fitting in limma. Raw P-values were adjusted for multiple comparisons by the Benjamini-Hochberg procedure [39] which controls the False Discovery Rate. We considered genes with an adjusted P-value,0.05 to be DE. Moreover, expression ratios, indicating the fold change in expression between two samples, were calculated using 20 h sample as a reference. Only genes with a |fold change (log 2 )| greater than 1 (standing for an increase or decrease in expression of at least 2 times) were considered as biologically significant. Genes fitting both P-value and fold change criteria were retained. The data presented in this publication have been deposited in NCBIs Gene Expression Omnibus (GEO, http://www.ncbi.nlm. nih.gov/geo/) and are accessible though GEO Series accession number GSE30841.
The DE genes were also classified according to their temporal expression profiles using the quadratic regression method proposed by Liu et al. [14].

Reverse transcription quantitative PCR
In order to confirm the results from the transcriptomic analyses, reverse transcription quantitative PCR (RT-qPCR) experiments were carried out. Transcripts were quantified at two specific times, at the exponential phase of growth (reference time 20 h) and after a 3-day incubation at 4uC. Twenty-eight genes observed as DE using microarray experiments were chosen. Primers were designed using Primer3Plus software [40] with default parameters except for the difference of melting temperature between the forward and reverse primers set to less than 1uC. The primer sequences for the tested genes are detailed in Table 6. cDNA synthesis, quantitative PCR, and cycle thresholds (Ct) were performed according to Falentin et al. [1]. Briefly, cDNA was synthesized using a qScript TM cDNA synthesis kit (Quanta BioSciences, Maryland, USA). Amplification by qPCR was performed with a 15 ml final volume mixture containing 5 ml of a cDNA template dilution of 1 in 40, 0.5 mM of each primer and 16 IQ TM SYBR Green Supermix (BioRad, California, USA), in an Opticon 2 real-time PCR detector (Biorad). Amplification cycles consisted in an initial step at 95uC for 5 min followed by 40 cycles of 95uC for 30 s, 60uC for 30 s, and 72uC for 1 min. Amplicon denaturation step of 0.5uC/min from 65uC to 90uC was performed to verify amplification specificity and determine amplicon melting temperature. Five genes that were not DE in microarray experiments (PFREUD_18870, gyrB1, 16S, pf279, gtfF) were chosen as possible internal standards for RT-qPCR normalization. The stability of mRNA expression of these genes was checked by using the geNorm VBA applet for Microsoft Excel [41]. Thus, PFREUD_18870, pf279 and gtfF genes were chosen as internal standard for normalization. The delta-delta-Ct method of geNorm was used to determine the normalized expression level of genes of interest (http://medgen.ugent.be/genorm/). An ANOVA was performed and changes in gene expression between the two conditions with a P-value,0.05 were considered as significant. To facilitate comparison with transcriptomic data, the results of RT-qPCR were expressed as fold change in log 2 .

Protein extraction
Protein extraction was performed as described by Jan et al. [42], with modifications. In brief, cells were harvested at four sampling times (20 h and 3, 6 and 9 days) by centrifugating 50 ml of culture (70006g, 10 min, room temperature or 4uC regarding the sampling time) and washed in a wash buffer (50 mM Tris-HCl The lysates were centrifuged (100006g, 10 min, room temperature) and the supernatants were stored at 220uC until electrophoresis.

Two-dimensional electrophoresis (2D electrophoresis) and image analysis
Proteins were precipitated (2D clean up kit, GE Healthcare) and rehydrated in 100 ml of Destreak rehydratation solution (GE Healthcare) containing 20% IPG buffer (pH 4/7, GE Healthcare). Isoelectric focusing was carried out using pH 4 to 7 Immobiline Dry Strips (18 cm, GE Healthcare) on a Multiphor II electrophoresis system (Amersham Biosciences). Second dimension SDS PAGE (14% monomers) was performed on an Ethan Dalt Twelve (GE Healthcare) according to a procedure previously described [42]. Analysis of the gel images was undertaken using Progenesis SameSpots software version 3.3 (Nonlinear Dynamics, Newcastle upon Tyne, United Kingdom). The gel images were aligned by automated calculation of alignment vectors after assigning 30-35 landmark vectors [43]. An ANOVA was performed and a threshold of P,0.05 was chosen to identify significant changes in spot volumes. Differentially expressed proteins also have to match a fold change criterion, expressed in log 2 , of at least 0.6.

Nano-LC/ESI-MS-MS analysis
Protein identification was performed using tandem mass spectrometry (MS/MS) as previously described [44]. All mass spectrometry data were submitted to MASCOT v.2.2 software for search into two concatenated databases: (i) Swiss-Prot database (http://www.expasy.org) and (ii) a portion of the UniProtKB database corresponding to the P. freudenreichii taxonomic group. Fifty protein spots were excised and submitted to an in-gel tryptic digestion as previously described [18]. The alkylation of cysteine by iodoacetamide was specified as a fixed modification while the two following modifications were set as variable modifications: oxidation of methionine and deamidation of asparagine and glutamine residues. The peptide mass tolerance was set to 0.2 Da for MS 0.15 Da for MS/MS. For each peptide identified, a minimum of four peptides with MASCOT score corresponding to a P-value,0.05 or an Exponentially Modified Protein Abundance Index (emPAI) [45] greater than 0.4 were necessary for validation with a high degree of confidence. For the validation of peptides from the MASCOT search results, the 1.19.2 version of the IRMa software was used [46].

Analysis of metabolites in culture supernatants
Free and total amino acids were analyzed with an amino acid analyzer (AlphaPlus series 2; Pharmacia, Uppsala, Sweden). Lactic, propionic, acetic, succinic, and pyruvic acids were quantified in culture supernatants by HPLC on an Aminex A-6 ion exchange column (Bio-Rad, Hercules, CA) at 55uC with 0.005 M H 2 SO 4 as eluent at a flow rate of 0.4 ml.min 21 . Both UV (210 nm) and refractometric detectors were used. Methylbutyric acid (sum of 2-methylbutyric and 3-methylbutyric acids) concentration was determined in culture supernatants after acidification by oxalic acid (final concentration: 0.03 mM) by gas chromatography, as previously described [12]. The rate of metabolites consumption or production was calculated i) at 30uC, between 20 h and 40 h, ii) at 4uC, by linear regression of values at lactate addition time point (64 h), at 3, 6 and 9 days. The differences between means from independent triplicate cultures for concentrations, ratios, and rates of metabolite consumption or production at the different sampling times were assessed by Student tests.   Figure S2 An illustration of the two main temporal expression patterns of P. freudenreichii CIRM-BIA1 T strain genes identified by the quadratic regression method proposed by Liu et al. [15]. The black dots are the hybridization signals. The curve is the fitted regression pattern. Pattern and genes given as an example of the pattern are as follows A: example of a quadratic linear concave up regulated regression pattern (for PFREUD_06720 gene) displayed by 178 differentially expressed genes; B: example of a quadratic linear convex down regulated regression pattern (for groEL1 gene) displayed by 272 differentially expressed genes.

Supporting Information
(TIF) Table S1 Concentrations of total and free amino acids, and NH 3 in P. freudenreichii CIRM-BIA1 T strain supernatant samples over time. (DOC)