Microorganisms have, for millennia, played a central role in the discovery and development of fermented food by humans. It has been observed that the biogeography of microorganisms is influenced by human practices, as microorganisms have been domesticated to different food matrices that are produced in different regions (Legras et al., 2007; Douglas and Klaenhammer, 2010). Even for foods and beverages that are made nearly worldwide such as bread and wine, in which species are not always specific to a region or product, local variations in the biogeography of microorganisms have been observed in the form of genomic traces (Legras et al., 2007; Almeida et al., 2015). Even if Saccharomyces cerevisiae is the main yeast species responsible for the fermentation of wine, the contribution of the microbiological signature of non-Saccharomyces genera to the development of typical wine aroma has already been studied (Capozzi et al., 2015). The complexity of the ecosystem associated to oenological environments leads to a discussion about the possible existence of the so-called microbial terroir (Gilbert et al., 2014). Evidence suggests, at least for wine, that soil microbiome influences the grapevine-associated microbiota and that this microbial signature might be partially responsible for differential wine phenotypes (Bokulich et al., 2014; Zarraonaindia et al., 2015; Knight et al., 2015).

Oenococcus oeni is the main species responsible for the malolactic fermentation (MLF) of wine, which normally follows the alcoholic fermentation (AF) produced by yeasts (Davis et al., 1986). Multilocus sequence typing analyses have revealed that the population of O. oeni is composed of a great diversity of strains that cluster in at least three main genetic groups, named A, B and C and a number of subgroups that sometimes correlate with their region (Chile, South Africa, Eastern France) or product of origin (cider, red and white wines, champagne) (Bilhère et al., 2009; Bridier et al., 2010). Genomic studies based on sequences of a few tens of strains have confirmed these genetic groups and they have suggested that ancestral O. oeni strains associated with fruits have been progressively domesticated to develop in cider and in wine, the strains of group A being presumably the best-adapted to wine (Borneman et al., 2012; Campbell-Sills et al., 2015). Recently, analysis of nearly 200 genomes, mostly from Australian isolates, showed that more than 60% of Australian isolates cluster in a closely-related group (a subgroup of A), suggesting that strains of this group may out-compete the other strains during fermentation or that they are well suited to Australian winemaking conditions (Sternes and Borneman, 2016).

During MLF O. oeni converts malic acid into lactic acid and CO₂, which makes wine softer in taste (Lonvaud-Funel, 1999). It also produces or degrades numerous secondary metabolites that can modify the fruity, vegetal or smoked aromas and contribute to the overall complexity of wine aroma (de Revel, 1999; Bartowsky, 2005; Vallet et al., 2008; Antalick et al., 2010; Antalick et al., 2012). Several studies have been made regarding the impact of different strains of O. oeni and other LAB in the composition of wine after MLF, both in primary and secondary metabolites (Pozo-Gayón et al., 2005; Ugliano and Moio, 2005; Lee et al., 2009a; Lee et al., 2009b; Hernandez-Orte et al., 2009; Costello et al., 2013; Sumby et al., 2013; Malherbe et al., 2013). However, it is still unknown whether strains of the same group have similar impacts on the quality of wines.

In a recent survey of lactic acid bacteria present during MLF in wines and ciders produced in diverse regions of France, we have collected nearly 3000 isolates of O. oeni, whose genotyping revealed 514 strains clustered into 43 genetic groups (El Khoury et al., 2017). Most groups contain strains collected in different regions or different types of wines. Conversely, each regional wine, including those of Burgundy, contains strains that cluster in several different genetic groups. Nevertheless, two peculiar groups were detected from Burgundy wines. Group A5 contained 17 strains isolated almost exclusively from white wines of this region and group A2.8 included 28 strains from the region's red wines. The present study aimed at analysing the specificities of these groups. We have analysed strains of both groups at the genomic level in order to confirm their clustering and to determine their phylogenetic position within the O. oeni species. We have also compared their capacity to perform MLF in both types of wines and their impact on the volatile compounds. The results suggest that these two groups of strains have evolved to adapt to the two types of wines found in Burgundy and other regions and that the presence of strains from one or the other group during MLF modulates differently the quality of wines.

O. oeni strains were obtained from the Biological Resources Center Oenology (CRBO) of ISVV (Villenave d’Ornon, France). Strains CRBO_14194, CRBO_14195, CRBO_14196, CRBO_14198, CRBO_14200, CRBO_14202 and CRBO_14203 were isolated from Chardonnay wines of Burgundy and strains CRBO_14205, CRBO_14206, CRBO_14207, CRBO_14210, CRBO_14211, CRBO_14212 and CRBO_14213 from Pinot noir wines of Burgundy. Strain CRBO_11105 was isolated from a red wine of Aquitaine and strain CRBO_14214 from red wine of Val de Loire. All the strains were propagated at 26 °C in a grape juice medium containing 25% commercial grape juice, 5 g/L of yeast extract and 0.1% tween80, adjusted to pH 4.8 with KOH. Commercial strains PN4^TM and VP41^TM were obtained from Lallemand SAS in their commercial format.

MLF trials were performed in a Pinot noir (12.6% alcohol, pH 3.15, titratable acidity 4.90 g/L and malic acid 3.6 g/L) and a Chardonnay wine (12.8% alcohol, pH 3.02, titratable acidity 5.10 g/L and malic acid 3.1 g/L) from Burgundy region. They were filter sterilised progressively at 3 µm, 0.8 µm and 0.2 µm. Filtered wines were stocked in 70 mL tubes at 4 ºC until inoculation. Four strains of each group A5 and A2.8 were propagated in grape juice medium, collected by centrifugation and inoculated to 2·10⁶ cells/mL in wine to start MLF. Lyophilised commercial strains were used according to the manufacturer’s instructions (Lallemand SAS) and were inoculated at 2·10⁶ cells/mL. MLF were carried out at 20 °C in 20 mL flasks with a minimum of contact with air. Trials were performed in triplicate and MLF progression was followed twice per week in only one of the replicates in order to limit the contacts with air for the two other replicates. MLF progression was monitored by determining malate concentration using the Roche L-Malic acid kit according to the manufacturer’s recommendations (r-Biopharm).

Microbial DNAs used for genome sequencing were extracted using the wizard genomic DNA purification kit according to manufacturer’s recommendations (Promega). The genomic DNAs were sequenced by Illumina MiSeq technology with paired-end reads and read length of 250bp at the Genomic and Transcriptomic Facility of Bordeaux. The obtained reads were cleaned with trim_galore v. 0.4.0 and extended with FLASH v1.2.11 (Magoc and Salzberg, 2011). Genomes were assembled de novo with Minia v. 1.0.6 (Chikhi and Rizk, 2013). Each genome was assembled either from the clean reads, either from the clean and extended reads, with kmer lengths of 25, 37 and 49, giving a total of 6 independent assemblies per genome. Assembly statistics were calculated using homemade scripts and the best of the six assemblies for each genome was kept based on their assembly statistics (N50, N90, largest contig size, shortest contig size).

The distances between genomes were calculated using ANIm algorithm with JSpecies v. 1.2.1 software (Richter and Rosselló-Móra, 2009). The obtained similarity matrix was transformed into a distance matrix and parsed into the format required by MEGA using homemade scripts. Phylogenomic trees were reconstructed by the neighbour joining method with MEGA v. 6.06 (Tamura et al., 2013).

Genomes were annotated on the RAST platform with Classic RAST annotation scheme, RAST gene caller and FIGfam Release70 (Aziz et al., 2008). Frame shifts fixing was turned on. The features of the genomes annotated by RAST were systematically classified in subsystems as part of the annotation pipeline and data mining was facilitated through the SEED environment (Overbeek et al., 2014). A matrix containing the quantity of genomic features classed into each subsystem category was built for each strain. For cluster analysis, the matrix was normalised with the formula log1p(x-min(x)), where x represents the number of features. The clusterisation was performed using Canberra distances and Ward clustering method using pheatmap R package. Since Canberra distances computation does not admit vectors composed of only 0’s, the normalised categories composed of only 0’s were replaced by 1’s; it doesn’t have any effect in the clusterisation given that they represent non-informative categories (i.e. all the strains have the same number of features for the same category, hence they do not contribute to their discrimination).

Ethanol (≥99.9%) was obtained from Merck (Damstadt, Germany). Ethyl butyrate-4,4,4-d₃ (>99%), ethyl hexanoate-d₁₁ (>98%), ethyl octanoate-d₁₅ (>98%) and ethyl trans-cinnamate-d₅ (phenyl-d₅) (>99%) were obtained from Cluzeau (Sainte Foy la Grande, France). Sodium chloride (norma pure) was purchased from VWR Chemicals (Fontenay-sous-Bois, France).

A method was adapted from Antalick et al. (2010). 5 μL of internal standard solution (ethyl butyrate-4,4,4-d₃, ethyl hexanoate-d₁₁, ethyl octanoate-d₁₅ and ethyl trans-cinnamate-d₅ (phenyl-d₅) at 200 mg/L in ethanol) was added to 5 mL of wine, then introduced into a 20 mL standard headspace vial filled with 3.5 g of sodium chloride. The solution was homogenized with a vortex shaker and then loaded onto a Gerstel autosampling device. The program consisted of swirling the vial at 500 rpm for 2 min at 40 °C, then inserting the fibre into the headspace for 30 min at 40 °C as the solution was swirled again, then transferring the fibre to the injector for desorption at 250°C for 15 min. The fibre used was polydimethylsiloxane 100 µm (PDMS-100) (Supelco, Bellefonte, PA, USA). It was conditioned before use as recommended by the manufacturer.

Gas chromatographic analyses were carried out on an Agilent 7890A GC system coupled to an Agilent 5975C quadrupole mass spectrometer and equipped with a Gerstel MPS2 autosampler. Injections were in the splitless mode for 0.75 min, using a 2 mm I.D. non-deactivated direct liner. A BP21 capillary column (50 m x 0.32 mm, 0.25 µm film thickness, SGE, Courtaboeuf, France) was used and the carrier gas was helium N55 with a column-head pressure of 8 psi. The oven temperature was programmed at 40 °C for 5 min then raised to 220 °C at 3 °C/min and then held at that temperature for 30 min. The mass spectrometer was operated in electron ionization mode at 70 eV with selected-ion-monitoring (SIM) and SCAN mode.

All raw chromatogram files were exported from Agilent Chemstation version D.03.00.611 (Agilent Technologies) as netCDF-files and imported into MATLAB version 8.0 (The MathWorks Inc., Natick, MA, USA) using built-in functions. In-house written and PLS-Toolbox functions have been used for further data processing in MATLAB. Preprocessing of the multi-way array was done using the nprocess.m function of the N-way toolbox. Prior to the mathematical transformations useless parts of the chromatogram at the beginning and at the end were removed. The data analysis approach has been reported recently (Vestner et al., 2016). The methodology consists of the segmentation of full scan GC-MS chromatograms along the retention time axis (corrected by an internal standard) and mathematical transformations including the calculation of sums of squares and cross product (SSCP) matrices of segments. The result of the segmentation and mathematical transformation is a three-way array with the dimensions number of samples × number of samples × number of segments (first and second mode are identical) which can be decomposed using parallel factor analysis (PARAFAC). Loadings of the first and second mode (sample mode) of the PARAFAC model can be interpreted in the same way as PCA scores, while the loadings of the third mode (segment mode) are represented as congruence loadings which represent the contribution (‘correlation’) of a segment on the corresponding PARAFAC component. Segments with high congruence loadings (> 0.75) are considered to ‘highly correlate’ with the corresponding component and therefore, as important to explain systematic differences among samples which are represented by this component in the sample mode loadings (‘scores’). Important segments are deconvoluted and peak profiles are integrated using AMDIS (Stein, 1999) and corrected by an internal standard. All peaks that are significantly different (Student’s t-test, alpha = 0.5) between the two groups of lactic acid bacteria are compiled in a peak table. The identification of peaks is done by comparing their spectra against the NIST database.

Six and eight strains of groups A5 (white wine) and A2.8 (red wine), respectively, were selected to produce and compare their genomes. They were sequenced by the Illumina method and assembled to produce drafts of 127 to 287 contigs (table 1).

All the reported genomes have a size of around 1.8 Mb, which is consistent with previous reports for O. oeni (Mills et al., 2005; Borneman et al., 2010; Borneman et al., 2012; Campbell-Sills et al., 2015). The number of protein encoding genes (PEG) that were detected and annotated by RAST fall in the order of ~1800, which is also comparable with data reported in the literature (Mills et al., 2005; Borneman et al., 2010; Borneman et al., 2012; Campbell-Sills et al., 2015; Sternes and Borneman, 2016). To determine their phylogenetic distribution, a phylogenomic tree was reconstructed with these 14 newly sequenced genomes and 50 additional ones reported on NCBI (Borneman et al., 2012, Campbell-Sills et al., 2015). The tree was calculated from ANIm distances and reconstructed by the neighbour joining method. Figure 1 confirms that all the new strains belong to the genetic group A as suggested previously (El Khoury et al., 2017). They cluster in two separate groups that most likely represent subgroups A5 and A2.8. Surprisingly, the tree also revealed that the two groups are closely related and that they are more distant from all other genomes. This suggests that they have diverged from a common ancestor strain fairly recently, at least more recently than their separation from all other groups of strains. In addition, four strains isolated from champagne (IOEB_B16, IOEB_0205, AWRIB422 and AWRIB548) cluster together with white wine strains of group A5 and group A2.8 has two strains isolated from red wines of Aquitaine (CRBO_11105) and Val de Loire (CRBO_14214). The clustering of these strains is not inconsistent because although they come from different regions, they were isolated from wines that have comparable physicochemical properties as those of the white and red wines of Burgundy.

Table

titre du tableau
Group	Strain	Sequence coverage (X)	Accession	Genome size (bp)	Number of contigs	N50	L50	N90	L90	PEGs
A5	CRBO_14194	38	SAMN04122363	1786610	196	27411	18	4263	82	1847
	CRBO_14195	71	SAMN04122364	1789621	127	49436	13	7354	49	1853
	CRBO_14196	48	SAMN04122365	1798795	208	27547	23	5901	77	1862
	CRBO_14198	88	SAMN04122366	1789795	174	28822	19	6019	73	1850
	CRBO_14200	90	SAMN04122367	1789801	167	39836	13	5457	64	1847
	CRBO_14203	48	SAMN04122369	1807672	131	40244	15	7105	55	1874
A2.8	CRBO_11105	48	SAMN04122350	1793882	200	28533	23	4638	81	1830
	CRBO_14205	66	SAMN04122370	1729210	225	23427	21	3884	93	1772
	CRBO_14206	63	SAMN04122371	1738384	202	25660	21	4438	86	1790
	CRBO_14207	40	SAMN04122372	1779011	251	24022	23	4989	81	1806
	CRBO_14210	64	SAMN04122373	1830066	202	28303	19	5172	81	1893
	CRBO_14211	46	SAMN04122374	1775057	287	13491	39	3274	139	1822
	CRBO_14213	102	SAMN04122375	1814591	137	38947	15	7291	55	1901
	CRBO_14214	50	SAMN04122376	1754584	271	15632	33	3074	130	1786

Table

Strains of both groups were used to carry out MLF in a red wine of Pinot noir and a white wine of Chardonnay in order to determine if the resulting wines differ according to the type of strains. Four strains from each group were propagated in a laboratory medium and inoculated in wines to 2·10⁶ cells/mL. Two commercial strains were used as positive control and a non-inoculated wine used as negative control. MLF was monitored during sixty days after inoculation. Strains of group A2.8 could achieve MLF in the Pinot noir wine with performances comparable to that of commercial strains, but A5 strains failed to start MLF in the same matrix (not shown). It is likely that these strains have a low tolerance for red wine as they also failed to start MLF in other trials performed in Pinot noir and Gamay wines (not shown). In the Chardonnay wine, group A5 strains were able to completely deplete malic acid (table 2) and so did the commercial strains used as positive control, while A2.8 strains had an heterogeneous behaviour.

Table

titre du tableau
Group	Strain-repetition	Malic acid (mg/L)
A5	CRBO_14194-A	<0.01
	CRBO_14194-B	<0.01
	CRBO_14195-A	<0.01
	CRBO_14195-B	<0.01
	CRBO_14196-A	<0.01
	CRBO_14196-B	<0.01
	CRBO_14202-A	<0.01
	CRBO_14202-B	<0.01
A2.8	CRBO_14206-A	<0.01
	CRBO_14206-B	<0.01
	CRBO_14208-A	0.553
	CRBO_14208-B	0.52
	CRBO_14210-A	0.066
	CRBO_14210-B	1.597
	CRBO_14212-A	1.805
	CRBO_14212-B	2.051
Commercial	PN4™-A	0.009
	PN4™-B	0.137
	Lalvin VP41™-A	<0.01
	Lalvin VP41™-B	<0.01
Control	Control-A	2.363
	Control-B	2.24

The A2.8-group strains had an opposite behaviour. While they all completed MLF in red wine, only one completely degraded malic acid in the white wine and the other three had consumed only 10 to 80%. After the same period, malic acid was not degraded in a non-inoculated control wine.

Since strains of group A5 failed to initiate MLF in red wine, the comparison of volatile compounds produced by the strains of the two groups during MLF was only performed in white wines. The resulting MLF samples were characterised by HS-SPME-GC and analysed under an untargeted metabolomics pipeline based on a PARAFAC model following the method proposed by Vestner et al. (2016). One sample (CRBO_14212A) was discarded since it represented an outlier according to the algorithm of Filzmoser, Maronna and Werner. Segmentation of the chromatograms resulted in a total of 86 segments. 24 segments containing only baseline or artefact peaks such as siloxane peaks from column bleeding were excluded from the data set. The three-way array obtained from mathematical transformations of the remaining 61 segments had the dimensions 19 × 19 × 61 (number of samples × number of samples × number of segments) including duplicates of each sample. PARAFAC models with 2 to 15 components were built to examine the optimal number of components. Core consistency diagnostic, residuals, captured variance and interpretability of loadings were examined to find an appropriate PARAFAC model which explains the variation among samples the best. An 11 component PARAFAC model gave the best interpretable results by explaining 81.6 % of the total variation in the dataset. PARAFAC components two (16.1 % explained variation), seven (6.6 % explained variation) and eleven (4.5% explained variation) contain information on systematic differences between the two groups of samples (figure 3), while the other components reflect only unsystematic differences in the chromatograms. The segments 48 and 57 on component 2, the segments 15, 23 and 39 on component 7 and the segments 23 and 39 on component 11 are responsible for the differentiation of the two groups of samples. These segments are considered to be ‘highly correlated’ with the raw data (congruence loadings > 0.75). Only peaks from these 5 segments were deconvoluted and integrated using AMDIS. All integrated peaks were checked for differences between mean values of the two groups of samples using Student’s t-test with alpha = 5%. Five peaks showed significant differences between the two groups of samples.

Table

Two modes of PARAFAC are superposed: the samples mode (dots) and the loadings mode (arrows). The colours of the points and polygons indicate the group of the strains, either ‘white wine’ (yellow), ‘red wine’ (red) or commercial strains (black).

In this study we delved into the biological and oenological significance of a monophyletic clade composed of two groups of O. oeni strains previously detected in red and white wines of Burgundy, throughout a genomics/metabolomics analysis. The two groups do not contain all the strains that develop in the red and white wines produced in this region, but they are remarkable because each of them contains strains that have been isolated from a single type of wine, suggesting that they are specifically adapted to develop either in red or in white wines (El Khoury et al., 2017). Our phylogenomics analysis of 14 new genome sequences produced from these strains confirms their clustering and also reveals two unanticipated features. First, strains derived from champagne wines cluster together with strains of group A5 isolated from white wines. The group of champagne strains was previously detected by multi-locus sequence typing (Bridier et al., 2010) and comparative genomics (Campbell-Sills et al., 2015). Although it was initially thought that these strains were specific to champagne wines, their new position in group A5 clearly indicates that they are present in wines of different regions. Nevertheless, champagne wines are white wines and they share some physicochemical properties with those produced in Burgundy, especially a low pH. Therefore, it is not surprising that these strains group together. On the contrary, this supports the hypothesis that group A5 strains are specifically adapted to this type of wine. The second new information revealed by the phylogenomic analysis is that the two groups of strains A5 and A2.8 are phylogenetically very close. This was unanticipated because all the O. oeni strains sequenced to date split into a large number of lineages which are more or less close to each other. The most distant wine strains reported to date are those of groups A and B, but even in group A the diversity is very important (Campbell-Sills et al., 2015; Sternes and Borneman, 2016). Therefore the close proximity of two groups of strains that are supposedly associated with either red or white wines might have a special significance. It is likely that these two groups derive from a common ancestor and it is tempting to speculate that they split in the region of Burgundy, since the vast majority of strains were isolated from wines of this region (El Khoury et al., 2017). However, a different scenario is possible. The two groups could have split in another region and, because the strains have capacities to develop in white or red wines, they were able to colonise those produced in Burgundy.

Although the objective of this study was not to compare the fermentation capacities of the two groups of bacteria, they were tested in different types of wines (red and white) to perform MLF. The strains of group A5 proved unable of initiating MLF in the red wine, whereas they performed well in white wine and conversely, those in group A2.8 achieved MLF in the red wine, but not always completely in the white wine. The results suggest that the origin of the strains correlates well with their fermentation capacities. However, the wines used in these trials were sterile filtered before inoculating the bacteria, which is not normal winemaking conditions and could have made the wines more difficult to ferment. Interestingly, even if the strains performed MLF in the white wine more or less efficiently, they produced wines whose volatile fractions were different according to the group of strains. Previous studies have shown that different strains have different impacts on the volatile fraction of wines (Pozo-Gayón et al., 2005; Ugliano and Moio, 2005; Lee et al., 2009a; Lee et al., 2009b; Hernandez-Orte et al., 2009; Costello et al., 2013; Sumby et al., 2013; Malherbe et al., 2013), but this is the first time that this impact is correlated to the genetic proximity of the strains. The fact that the two groups of strains show differences in the volatile fraction of the fermented wines shed a new light on the existence of microbiological component associated with given wines and on the possible repercussions of the highlighted microbial diversity on the typical quality traits of regional wines, which is a field of considerable economic importance (Capozzi and Spano, 2011).

It is still difficult and speculative to correlate the genetic differences observed between the genomes of the two groups of strains and their possible adaptation to a type of wine or their capacity to produce volatile compounds during MLF. Nevertheless, since the two groups of strains are phylogenetically close, the sub-systems analysis revealed a limited number of specific genes in each group of strains, whose functions are particularly interesting. For instance, white wine strains carry the fructose specific components of the PTS, while red wine strains have the mannitol specific components. The features of PTS provide bacteria a system to assure optimal utilisation of carbohydrates in complex environments (Kotrba et al., 2001) and variations in the PTS enzyme II sugar transporters have already been observed for a large collection of O. oeni strains (Sternes and Borneman, 2016). Several sugars are present in wine after alcoholic fermentation, especially fructose and pentoses such as ribose, arabinose and xylose (Ribéreau-Gayon et al., 2012). LAB can use fructose as an e^- acceptor to produce mannitol during heterolactic fermentation, which permits the generation of ATP (Hornsey, 2007; Lahtinen et al., 2011). It has been reported that O. oeni can use the mannitol pathway in fructose fermentation due to limiting redox regeneration capacity of the ethanol pathway and that the choice of the fermentation pathway between mannitol and fructose is tightly regulated in O. oeni in order to maintain the equilibrium of NAD(P)H (Richter et al., 2003a; Richter et al., 2003b; Cibrario et al., 2016). It is not surprising then that the presence of the mannitol specific PTS components in red wine strains correlate with the presence of genes of oxidative stress response, as there are specific stressors characterizing red wines with respect of white ones. This is not the only function found in this study that might be related to the stress adaptation of O. oeni: a Dps protein that is lost in white wine strains, but present in red wine strains, has been observed to correlate with fitness in red wine (Bon et al., 2009). In effect, another study has shown that E. coli over-expressing this gene has gained resistance to wine, copper and ferric ions (Athané et al., 2008).

Exopolysaccharydes are very important for the adaptation of O. oeni to its ecological niche (Dimopoulos et al., 2014). All the white wine strains carry the gtf gene, which is absent in all the red wine strains. The presence of this gene is correlated to an increased resistance to several stresses occurring in wine (alcohol, pH, SO₂) (Dols-Lafargue et al., 2008). In particular, among this stressors, in the case of champagne and white wines of Burgundy the acidity is higher when compared to other wines. In the study by Dols-Lafargue et al. (2008), 7 out of 8 strains carrying the gtf gene had been isolated from white wine or Champagne. Just as for the genes of sugar utilisation, the presence of the gtf gene is not only a matter of survival for O. oeni, but might also have consequences at the organoleptic level since it is sometimes associated to a ropiness phenotype in wine (Dols-Lafargue et al., 2008; Dimopoulos et al., 2014).

Succinate and its derived esters are normally present in wine (Ribéreau-Gayon et al., 2012). The formation of diethyl succinate during MLF carried out by O. oeni has been reported several times (Pozo-Bayón et al., 2005; Ugliano et al., 2005; Izquierdo Cañas et al., 2008). Succinate, one of the precursors of diethyl succinate, can be combined with L-homoserine by the enzyme homoserine O-succinyltransferase (HSST), coded by the gene metA, in the reversible reaction succinyl-CoA + L-homoserine ⇌ CoA+ O-succinyl-L-homoserine. The HSST enzyme is also the first step in one of the three possible pathways of L-methionine biosynthesis from L-homoserine (Liu et al., 2008), with succinate being re-released in one of the intermediary reactions catalysed by the enzyme Cystathionine gamma synthase (CGS) (Rowbury and Woods, 1964a; Rowbury and Woods, 1964b; Liu et al., 2008). Although O. oeni does not carry the CGS enzyme, it does carry the cystathionine gamma lyase (CGL) enzyme, that has been reported to be able to produce α-ketobutyrate and succinate from O-succinyl-L-homoserine (Knoll et al., 2011). The transcription of the gene coding for HSST is repressed by L-methionine (Saint-Girons et al., 1988). A comparison against the genomes reported in Campbell-Sills et al. (2015) shows that all the white wine strains carry a frameshift mutation on the metA gene coding for the HSST enzyme, probably inactivating it. The gene coding the CGL enzyme, in exchange, is intact in all the strains. Our results suggest a link between the mutation of this enzyme in all the strains from white wine and the low levels of diethyl succinate produced, although the exact mechanism remains unknown. The fact that white wine strains could achieve MLF suggests that they are most probably obtaining L-methionine by other means; this is not surprising, since previous studies on 4 O. oeni strains determined that they were auxotroph for methionine (Remize et al., 2006).

This preliminary exploration of the genetic features of these groups of strains probably explains only part of their adaptability to white and red wines, as well as their differences in the production of volatile compounds during MLF. Further genomics and gene expression studies are in progress with the aim of identifying all their differences. Nevertheless, these results show that the diversity of strains encountered in different wines and regions is not entirely random. There are strains lineages that are genetically better suited to develop and perform MLF in certain types of wines, as they possess different enzymatic equipment that impacts on the volatile fraction of wines.

The study of two genetic groups of O. oeni strains associated with wines of Burgundy and other regions, throughout a genomics/metabolomics analysis offers biological insights on the possible genetic determinants of O. oeni adaptation to white and red wine associated environments, confirming the increasing interest in the examination of microbial diversity associated with fermented foods as possible general models in microbiology. Furthermore, we shed a new light on the existence of microbiological component associated with a given terroir and on the possible implications on the typical quality traits of regional wines. Further studies, including other non-volatile important metabolites and more strains of distant genetic groups, will give more clues on the impact of these variations at the organoleptic quality of wine.

This work was supported in parts by the European commission (FP7-SME project Wildwine, grant agreement n°315065) and the French Ministry of Agriculture (project CASDAR LevainsBio 2012-1220). Genomic DNA sequencing was performed at the Genomic and Transcriptomic Facility of Bordeaux (grants from the Conseil Regional d’Aquitaine n°20030304002FA and 20040305003FA and from the European Union, FEDER n°2003227 and from Investissements d'avenir, Convention attributive d’aide N°ANR-10-EQPX-16-01).

1 Almeida P. , Barbosa R. , Zalar P. , Imanishi Y. , Shimizu K. , Turchetti B. , Legras J.-L. , Serra M. , Dequin S. , Couloux A. , Guy J. Bensasson D. , Gonçalves P. and Sampaio J.P. (2015) . A population genomics insight into the Mediterranean origins of wine yeast domestication. Molecular Ecology 24, –. doi:10./mec.1. 2 Antalick G. (2010) . Bilan biochimique et sensoriel des modifications de la note fruitée des vins rouges lors de la fermentation malolactique : rôle particulier des esters (Doctoral Dissertation). Retrieved from theses.fr (Accession no. BOR2). 3 Antalick G. , Perello M.-C. and de Revel G. (2010) . Development, validation and application of a specific method for the quantitative determination of wine esters by headspace-solid-phase microextraction-gas chromatography–mass spectrometry. Food Chemistry 121, –. doi:10./j.foodchem..01.011. 4 Antalick G. , Perello M.-C. and de Revel G. (2012) . Characterization of fruity aroma modifications in red wines during malolactic fermentation. Journal of Agricultural and Food Chemistry 60, 1–3. doi:10./jf38n. 5 Athané A. , Bilhère E. , Bon E. , Morel G. , Lucas P. , Lonvaud A. and Le Marrec C. (2008) . Characterization of an acquired dps -containing gene island in the lactic acid bacterium Oenococcus oeni. Journal of Applied Microbiology 105, –. doi:10./j.-..7.x. 6 Aziz R.K. , Bartels D. , Best A.A. , DeJongh M. , Disz T. , Edwards R.A. , Formsma K. , Gerdes S. , Glass E.M. and Kubal M. et al. (2008) . The RAST Server: Rapid Annotations using Subsystems Technology. BMC Genomics 9, 75. doi:10./--9-75. 7 Bartowsky E.J. (2005) . Oenococcus oeni and malolactic fermentation–moving into the molecular arena. Australian Journal of Grape and Wine Research 11, 174–187. doi:10./j.-..tb6.x. 8 Bilhère E. , Lucas P.M. , Claisse O. and Lonvaud-Funel A. (2009) . Multilocus sequence typing of Oenococcus oeni: detection of two subpopulations shaped by intergenic recombination. Applied and Environmental Microbiology 75, –. doi:10./AEM.3-08. 9 Bokulich N.A. , Thorngate J.H. , Richardson P.M. and Mills D.A. (2014) . PNAS Plus: From the Cover: microbial biogeography of wine grapes is conditioned by cultivar, vintage and climate. Proceedings of the National Academy of Sciences 111, E139–E148. doi:10./pnas.10. 10 Bon E. , Delaherche , A. , Bilhere E. , De Daruvar A. , Lonvaud-Funel A. and Le Marrec C. (2009) . Oenococcus oeni genome plasticity is associated with fitness. Applied and Environmental Microbiology 75, –. doi:10./AEM.4-08. 11 Borneman A.R. , Bartowsky E.J. , McCarthy J. and Chambers P.J. (2010) . Genotypic diversity in Oenococcus oeni by high-density microarray comparative genome hybridization and whole genome sequencing. Applied Microbiology and Biotechnology 86, 681–691. doi:10./s3-009--6. 12 Borneman A.R. , McCarthy J.M. , Chambers P.J. and Bartowsky E.J. (2012) . Comparative analysis of the Oenococcus oeni pan genome reveals genetic diversity in industrially-relevant pathways. BMC Genomics 13, 373. doi:10./--13-373. 13 Bridier J. , Claisse O. , Coton M. , Coton E. and Lonvaud-Funel A. (2010) . Evidence of distinct populations and specific subpopulations within the species Oenococcus oeni. Applied and Environmental Microbiology 76, –. doi:10./AEM.4-10. 14 Campbell-Sills H. , El Khoury M. , Favier M. , Romano A. , Biasioli F. , Spano G. , Sherman D.J. , Bouchez O. , Coton E. and Coton M. et al. (2015) . Phylogenomic analysis of Oenococcus oeni reveals specific domestication of strains to cider and wines. Genome Biology and Evolution 7, –. doi:10./gbe/evv084. 15 Capozzi V. and Spano G. (2011) . Food microbial biodiversity and “microbes of protected origin”. Frontiers in Microbiology 2, 237. doi:10./fmicb..7. 16 Capozzi V. , Garofalo C. , Chiriatti M.A. , Grieco F. and Spano G. (2015) . Microbial terroir and food innovation: The case of yeast biodiversity in wine. Microbiological Research 181, 75–83. doi:10./j.micres..10.005. 17 Chikhi R. and Rizk G. (2013) . Space-efficient and exact de Bruijn graph representation based on a Bloom filter. Algorithms for Molecular Biology 8, 1. doi:10./--8-22. 18 Cibrario A. , Peanne C. , Lailheugue M. , Campbell-Sills H. and Dols-Lafargue M. (2016) . Carbohydrate metabolism in Oenococcus oeni: a genomic insight. BMC Genomics 17. doi:10./s4-016--2. 19 Costello P.J. , Siebert T.E. , Solomon M.R. and Bartowsky E.J. (2013) . Synthesis of fruity ethyl esters by acyl coenzyme A: alcohol acyltransferase and reverse esterase activities in Oenococcus oeni and Lactobacillus plantarum. Journal of Applied Microbiology 114, 797–806. doi:10./jam.8. 20 Davis C.R. , Wibowo D.J. , Lee T.H. and Fleet G.H. (1986) . Growth and metabolism of lactic acid bacteria during and after malolactic fermentation of wines at different pH. Applied and Environmental Microbiology 51, 539–545. 21 de Revel G. (1999) . Le diacétyle, les composés dicarbonyles et leurs produits de réduction dans le vin. Université de Bordeaux 2. 22 Dimopoulou M. , Vuillemin M. , Campbell-Sills H. , Lucas P.M. , Ballestra P. , Miot-Sertier C. , Favier M. , Coulon J. , Moine V. , Doco T. , Roques M. , Williams P. , Petrel M. , Gontier E. , Moulis C. , Remeaud-Simeon M. and Dols-Lafargue M. (2014) . Exopolysaccharide (EPS) synthesis by Oenococcus oeni: from genes to phenotypes. PLoS ONE 9, e8. doi:10./journal.pone.898. 23 Dols-Lafargue M. , Lee H.Y. , Le Marrec C. , Heyraud A. , Chambat G. and Lonvaud-Funel A. (2008) . Characterization of gtf, a glucosyltransferase gene in the genomes of Pediococcus parvulus and Oenococcus oeni, two bacterial species commonly found in wine. Applied and Environmental Microbiology 74, –. doi:10./AEM.3-08. 24 Douglas G.L. and Klaenhammer T.R. (2010) . Genomic evolution of domesticated microorganisms. Annual Review of Food Science and Technology - (new in ) 1, 397–414. doi:10./annurev.food.08.34. 25 El Khoury M. , Campbell-Sills H. , Salin F. , Guichoux E. , Claisse O. and Lucas P.M. (2017) . Biogeography of Oenococcus oeni reveals distinctive but nonspecific populations in wine-producing regions. Appl Environ Microbiol 83:e2-16. doi:10./AEM.2-16. 26 Gilbert J.A. , van der Lelie D. and Zarraonaindia I. (2014) . Microbial terroir for wine grapes. Proceedings of the National Academy of Sciences 111, 5–6. doi:10./pnas.10. 27 Hernandez-Orte P. , Cersosimo M. , Loscos N. , Cacho J. , Garcia-Moruno E. and Ferreira V. (2009) . Aroma development from non-floral grape precursors by wine lactic acid bacteria. Food Research International 42, 773–781. doi:10./j.foodres..02.023. 28 Hornsey I.S. (2007) . The Chemistry and Biology of Winemaking (The Royal Society of Chemistry). . 29 Izquierdo Cañas P.M. , García Romero E. , Gómez Alonso S. and Palop Herreros M.L.L. (2008) . Changes in the aromatic composition of Tempranillo wines during spontaneous malolactic fermentation. Journal of Food Composition and Analysis 21, 724–730. doi:10./j.jfca..12.005. 30 Knight S. , Klaere S. , Fedrizzi B. and Goddard M.R. (2015) . Regional microbial signatures positively correlate with differential wine phenotypes: evidence for a microbial aspect to terroir. Scientific Reports 5, 3. doi:10./srep3. 31 Knoll C. , du Toit M. , Schnell S. , Rauhut D. and Irmler S. (2011) . Cloning and characterisation of a cystathionine β/γ-lyase from two Oenococcus oenioenological strains. Applied Microbiology and Biotechnology 89, –. doi:10./s3-010--x. 32 Kotrba P. , Inui M. and Yukawa H. (2001) . Bacterial phosphotransferase system (PTS) in carbohydrate uptake and control of carbon metabolism. Journal of Bioscience and Bioengineering 92, 502–517. 10./S-(01)8. 33 Lahtinen S. , Ouwehand , A.C. , Salminen S. and von Wright A. (2011) . Lactic Acid Bacteria: Microbiological and Functional Aspects, Fourth Edition (Taylor & Francis). . 34 Lee J.-E. , Hwang G.-S. , Lee C.-H. and Hong Y.-S. (2009) . a. Metabolomics reveals alterations in both primary and secondary metabolites by wine bacteria. Journal of Agricultural and Food Chemistry 57, 2–3. doi:10./jf442. 35 Lee J.-E. , Hong Y.-S. and Lee C.-H. (2009) . b. Characterization of fermentative behaviors of lactic acid bacteria in grape wines through 1 H NMR- and GC-based metabolic profiling. Journal of Agricultural and Food Chemistry 57, –. doi:10./jf02a. 36 Legras J.-L. , Merdinoglu D. , Cornuet J.-M. and Karst F. (2007) . Bread, beer and wine: Saccharomyces cerevisiae diversity reflects human history. Molecular Ecology 16, –. doi:10./j.-294X..6.x. 37 Liu M. , Nauta A. , Francke C. and Siezen R.J. (2008) . Comparative genomics of enzymes in flavor-forming pathways from amino acids in lactic acid bacteria. Applied and Environmental Microbiology 74, –. doi:10./AEM.0-08. 38 Lonvaud-Funel A. (1999) . Lactic acid bacteria in the quality improvement and depreciation of wine. Ant. van Leeuwenhoek 317–331. doi:10./978-94-017--4_16. 39 Magoc T. and Salzberg S.L. (2011) . FLASH: fast length adjustment of short reads to improve genome assemblies. Bioinformatics 27, –. doi:10./bioinformatics/btr507. 40 Malherbe S. , Menichelli E. , du Toit M. , Tredoux , A. , Muller N. , Naes T. and Nieuwoudt H. (2013) . The relationships between consumer liking, sensory and chemical attributes of Vitis vinifera L. cv. Pinotage wines elaborated with different Oenococcus oeni starter cultures: Consumer liking, –. doi:10./jsfa, sensory and chemical attributes of Pinotage wines. Journal of the Science of Food and Agriculture 93. 41 Mills D. , Rawsthorne H. , Parker C. , Tamir D. and Makarova K. (2005) . Genomic analysis of Oenococcus oeni PSU-1 and its relevance to winemaking. FEMS Microbiology Reviews 29, 465–475. 42 Overbeek R. ,  Begley T. ,  M. Butler R. ,  V. Choudhuri J. ,   Chuang H.-Y. Cohoon M. , de Crécy-Lagard V. , Diaz N. , Disz T. , Edwards R. , Fonstein M. , Frank Ed. D. , Gerdes S. , Glass E.M. , Goesmann A. , Hanson A. ,  Iwata-Reuyl D. , Jensen Roy ,  Jamshidi N. , Krause L. , Kubal M. , Larsen N. ,  Linke B. , McHardy Alice C. Folker , Neuweger H. , Olsen G. ,  Olson R. ,  Osterman A. , Portnoy V. , D. Pusch G. , A. Rodionov D. ,  Rückert C. ,  Steiner J. , Stevens R. , Thiele I. , Vassieva O. ,  Ye Y. and Zagnitko O. and Vonstein V. (2005) . The subsystems approach to genome annotation and its use in the project to annotate genomes. Nucleic Acids Research 33, –. doi:10./nar/gki866. 43 Overbeek R. , Olson R. , Pusch G.D. , Olsen G.J. , Davis J.J. , Disz T. , Edwards R.A. , Gerdes S. , Parrello B. , Shukla M. , Vonstein V. , R. Wattam A. , Xia F. and Stevens R. (2014) . The SEED and the Rapid Annotation of microbial genomes using Subsystems Technology (RAST). Nucleic Acids Research 42, D206–D214. doi:10./nar/gkt. 44 Pozo-Bayón M.A. , G-Alegría E. , Polo M.C. , Tenorio C. , Martín-Álvarez P.J. , Calvo de la Banda M.T. , Ruiz-Larrea , F. and Moreno-Arribas M.V. (2005) . Wine volatile and amino acid composition after malolactic fermentation: effect of Oenococcus oeni and Lactobacillus plantarum starter cultures. Journal of Agricultural and Food Chemistry 53, –. doi:10./jf39y. 45 Remize F. , Gaudin , A. , Kong Y. , Guzzo J. , Alexandre H. , Krieger S. and Guilloux-Benatier M. (2006) . Oenococcus oeni preference for peptides: qualitative and quantitative analysis of nitrogen assimilation. Archives of Microbiology 185, 459–469. doi:10./s3-006--6. 46 Ribéreau-Gayon P. , Glories Y. , Maujean , A. and Dubourdieu D. (2012) . Traité d’œnologie - Tome 2 - 6e éd. - Chimie du vin. Stabilisation et traitements (Dunod). 47 Richter M. and Rosselló-Móra R. (2009) . Shifting the genomic gold standard for the prokaryotic species definition. Proceedings of the National Academy of Sciences 106, 6–1. doi:10./pnas.06. 48 Richter H. , De Graaf A.A. , Hamann I. and Unden G. (2003) . a. Significance of phosphoglucose isomerase for the shift between heterolactic and mannitol fermentation of fructose by Oenococcus oeni. Archives of Microbiology 180, 465–470. doi:10./s3-003--5. 49 Richter H. , Hamann I. and Unden G. (2003) . b. Use of the mannitol pathway in fructose fermentation of Oenococcus oeni due to limiting redox regeneration capacity of the ethanol pathway. Archives of Microbiology 179, 227–233. doi:10./s3-003--6. 50 Rowbury R.J. and Woods D.D. (1964) . a. O-Succinylhomoserine as an intermediate in the synthesis of cystathionine by Escherichia coli. Microbiology 36, 341–358. doi:10./-36-3-341. 51 Rowbury R.J. and Woods D.D. (1964) . b. Repression by methionine of cystathionase formation in Escherichia coli. Microbiology 35, 145–158. doi:10./-35-1-145. 52 Saint-Girons I. , Parsot C. , Zakin M.M. , Barzu O. and Cohen G.N. (1988) . Methionine biosynthesis in Enterobacteriaceae: biochemical, regulatory and evolutionary aspects. CRC Crit Rev Biochem 23 Suppl 1, S1–S42. doi:10./4. 53 Stein S.E. (1999) . An integrated method for spectrum extraction and compound identification from gas chromatography/mass spectrometry data. Journal of the American Society for Mass Spectrometry 10, 770–781. doi:10./S-(99)7-1. 54 Sternes P.R. and Borneman A.R. (2016) . Consensus pan-genome assembly of the specialised wine bacterium Oenococcus oeni. BMC Genomics 17.Stevens J.B., de Luca N.G., 228–231. . 55 Sumby K.M. , Jiranek V. and Grbin P.R. (2013) . Ester synthesis and hydrolysis in an aqueous environment and strain specific changes during malolactic fermentation in wine with Oenococcus oeni. Food Chem 141, –. doi:10./j.foodchem..03.087. 56 Tamura K. , Stecher G. , Peterson D. , Filipski A. and Kumar S. (2013) . MEGA6: Molecular Evolutionary Genetics Analysis Version 6. 0. Molecular Biology and Evolution 30, –. doi:10./molbev/mst197. 57 Ugliano M. and Moio L. (2005) . Changes in the concentration of yeast-derived volatile compounds of red wine during malolactic fermentation with four commercial starter cultures of Oenococcus oeni. Journal of Agricultural and Food Chemistry 53, 4–9. doi:10./jf672. 58 Vallet A. , Lucas P. , Lonvaud-Funel A. and de Revel G. (2008) . Pathways that produce volatile sulphur compounds from methionine in Oenococcus oeni. Journal of Applied Microbiology 104, –. doi:10./j.-..3.x. 59 Vestner J. , de Revel G. , Krieger-Weber S. , Rauhut D. , du Toit M. and de Villiers A. (2016) . Toward automated chromatographic fingerprinting: A non-alignment approach to gas chromatography mass spectrometry data. Analytica Chimica Acta 911, 42–58. doi:10./j.aca..01.020. 60 Zarraonaindia I. , Owens S.M. , Weisenhorn P. , West K. , Hampton-Marcell J. , Lax S. , Bokulich N.A. , Mills D.A. , Martin G. and Taghavi S. et al. (2015) . The soil microbiome influences grapevine-associated microbiota. mBio 6, e7–14. doi:10./mBio.7-14.