A methodological approach to screen diverse cheese-related bacteria for their ability to produce aroma compounds

lable at ScienceDirect

Food Microbiology 46 (2015) 145e153

Contents lists avai

Food Microbiology

journal homepage: www.elsevier .com/locate/ fm

A methodological approach to screen diverse cheese-related bacteriafor their ability to produce aroma compounds

Tomislav Poga�ci�c a, b, **, Marie-Bernadette Maillard a, b, Aur�elie Leclerc c,Christophe Herv�e c, Victoria Chuat a, b, Alyson L. Yee a, b, Florence Valence a, b,Anne Thierry a, b, *

a INRA, UMR1253 Science et Technologie du Lait et de l'Œuf, 65 rue de Saint Brieuc, 35000 Rennes, Franceb AGROCAMPUS OUEST, UMR1253 Science et Technologie du Lait et de l'Œuf, 65 rue de Saint Brieuc, 35000 Rennes, Francec Laboratoires Standa, F-14000 Caen, France

a r t i c l e i n f o

Article history:Received 6 May 2014Received in revised form8 July 2014Accepted 26 July 2014Available online 8 August 2014

Keywords:CheeseBacteriaAroma compoundsScreeningVolatile metabolite profilingVolatilome

* Corresponding author. INRA, UMR1253 Sciencel'Œuf, F-35042 Rennes, France. Tel.: þ33 223 485 337** Corresponding author. Present address: DepartmeAgriculture University of Zagreb, Sveto�simunska 2Tel.: þ385 1239 3646.

E-mail addresses: [email protected] (T. Poga�ci�c)(A. Thierry).

http://dx.doi.org/10.1016/j.fm.2014.07.0180740-0020/© 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

Microorganisms play an important role in the development of cheese flavor. The aim of this study was todevelop an approach to facilitate screening of various cheese-related bacteria for their ability to producearoma compounds. We combined i) curd-based slurry medium incubated under conditions mimickingcheese manufacturing and ripening, ii) powerful method of extraction of volatiles, headspace trap,coupled to gas chromatography-mass spectrometry (HS-trap-GC-MS), and iii) metabolomics-basedmethod of data processing using the XCMS package of R software and multivariate analysis. Thisapproach was applied to eleven species: five lactic acid bacteria (Leuconostoc lactis, Lactobacillus sakei,Lactobacillus paracasei, Lactobacillus fermentum, and Lactobacillus helveticus), four actinobacteria (Bra-chybacterium articum, Brachybacterium tyrofermentans, Brevibacterium aurantiacum, and Microbacteriumgubbeenense), Propionibacterium freudenreichii, and Hafnia alvei. All the strains grew, with maximalpopulations ranging from 7.4 to 9.2 log (CFU/mL). In total, 52 volatile aroma compounds were identified,of which 49 varied significantly in abundance between bacteria. Principal component analysis of volatileprofiles differentiated species by their ability to produce ethyl esters (associated with Brachybacteria),sulfur compounds and branched-chain alcohols (H. alvei), branched-chain acids (H. alvei, P. freudenreichiiand L. paracasei), diacetyl and related carbonyl compounds (M. gubbeenense and L. paracasei), amongothers.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Characterization of microorganisms for their production ofodor-active volatile compounds and evaluation of their utility asripening cultures in cheese manufacture is an ongoing scientificchallenge in dairy microbiology. New strains isolated from dairy ornon-dairy environments should be evaluated for their aromaticpotential, since flavor is a very important characteristic from theconsumer's point of view (Niimi et al., 2014). The formation offlavor compounds in cheese results from numerous metabolic

et Technologie du Lait et de; fax: þ33 223 485 350.nt of Dairy Science, Faculty of5, 10 000 Zagreb, Croatia.

, [email protected]

reactions and is largely influenced by microbial diversity and thecomplex dynamics of growth and metabolism during cheeseripening (Hassan et al., 2013; Steele et al., 2013). The microbiota oftraditional Protected Designation of Origin (PDO) raw milk cheesesdepends on the microbial community, which naturally arises fromrawmilk and natural whey culture, and from the environment, andcontributes to specific intense flavor of raw milk cheeses (Gattiet al., 2014; Neviani et al., 2013; Ordiales et al., 2013). However,in cheese manufacturing, there is a continual need to modulatecheese flavor via the addition of selected new strains with aromapotential, particularly in the case of cheeses made from pasteurizedmilk. For example, there is currently a demand to diversify therather mild flavor of some semi-hard cheeses. Therefore, efficientaroma screening approaches are required to evaluate diverse spe-cies of microorganisms. Most studies have targeted a few groups ofbacteria, mainly lactic acid bacteria (LAB) such as Lactobacillus,Lactococcus, and Leuconostoc, or propionibacteria (De Bok et al.,

T. Poga�ci�c et al. / Food Microbiology 46 (2015) 145e153146

2011; Dhaisne et al., 2013; Pedersen et al., 2013; Sgarbi et al., 2013;Zareba et al., 2014). Considering the diversity of the cheesemicrobiota, we aimed to evaluate the ability of diverse bacterialstrains (7 genera, 11 species), to grow and produce volatile aromacompounds in a unique curd-based medium under conditions thatmimicked those of semi-hard cheese manufacture.

Several methods can be used to analyze volatile compounds e

also referred to as the volatilome (Sgarbi et al., 2013 e (Croissantet al., 2011). Many volatile compounds have been identified incheese and other fermented dairy products (Maarse et al., 1994;Steele et al., 2013). They include free carboxylic acids, sulfur com-pounds, carbonyl compounds, alcohols, and esters, among others(Mariaca and Bosset, 1997; Curioni and Bosset, 2002). The analysisof volatiles involves sample preparation, extraction, and concen-tration, followed by separation, detection, and identification,generally performed by gas chromatography coupled to massspectrometry (GC-MS). The choice of the sample preparation andextraction method is crucial because it influences both the quali-tative and quantitative profiles of the volatiles extracted, as well asthe accuracy of the results (Marsili, 2011; Jele�n et al., 2012).Headspace (HS) sampling has become the most frequently usedtechnique in the investigation of food flavor. It is a means ofseparating volatiles from the sample prior to GC-MS analysis. HS-solid phase micro-extraction (SPME) is the most widely usedmethod to extract volatiles from food products, including cheese(Jele�n et al., 2012). Additionally, other HS-related micro-extractionmethods have been developed, for example, the HS-trap techniquehas recently been successfully applied to extract volatiles fromvarious samples (Schulz et al., 2007; Aberl and Coelhan, 2012).

GCMS data processing and analysis is the final step and can bevery time-consuming when there are many samples to compare. Agreat number of volatile compounds can potentially be extractedandmust be identified and quantified in each sample. Data analysiscan be facilitated using tools developed for MS-based metabolomicapproaches, which have emerged over the past decade in manyscientific areas, including food science (Wishart, 2008). To facilitatethe handling of large datasets generated from liquidchromatography/MS-based metabolomic approaches, dedicatedsoftware has been developed, including the open-source XCMSpackage (Smith et al., 2006). These tools convert the initial three-dimensional raw data (m/z, retention time, ion current) into atwo-dimensional data table containing information about theabundance of each metabolite in all samples (Antignac et al., 2011).

Our aim was to develop an efficient approach to screen variouscheese-related bacteria species for their ability to produce aroma

Table 1Bacterial strains used and conditions of revitalization and enumeration.

Straina Mediumb (broth or agar)

Brachybacterium articum LSBA53 TSB-YEBrachybacterium tyrofermentans LSBT17 TSB-YEBrevibacterium aurantiacum LSBA 57 TSB-YE þ NaClMicrobacterium gubbeenense LSMG39 TSB-YE þ NaClLactobacillus fermentum LSLF202 MRSLactobacillus helveticus CIRM-BIA108 MRSLactobacillus paracasei LSLP248 MRSLactobacillus sakei LSLS89 MRSLeuconostoc lactis CIRM-BIA1541 MRSPropionibacterium freudenreichii CIRM-BIA1426 YELHafnia alvei CIRM-BIA1620 BHI-YEHafnia alvei CIRM-BIA1621 BHI-YE

a Strain named CIRM-BIA are from CIRM-BIA, INRA, the other strains are from Laboratb Media: MRS: Man Rogosa Sharpe; TSB-YE: tripticase soy broth yeast extract; YEL: yea

agar.c Agitation at 90 RPM; AN, anaerobic; AE, air atmosphere.

compounds. Ideally, such an approach should allow assessment ofthe growth of a variety of targeted species, should be sensitive andsufficiently simple and automated to be useful for large-scalescreening. Our strategy was to combine the use of i) a curd-basedmedium incubated under conditions mimicking cheese manufac-ture and ripening, ii) the extraction of volatiles using the recentlydeveloped headspace trapmethod coupled to gas chromatography-mass spectrometry (GC-MS), and iii) a metabolomics-basedmethod of data processing using the open-source XCMS packageof R software, followed by statistical and multivariate analyses.

2. Materials and methods

2.1. Bacterial strains

Twelve bacterial strains (Table 1) of different species of interestfor cheese aromatization were used in this experiment, five fromthe collection of the International Centre for Bacteria of Food In-terest - Centre International de Ressources MicrobienneseBact�eriesd'Int�eret Alimentaire (CIRM-BIA, UMR1253, INRA Rennes, France)and seven from Laboratoires Standa, Caen, France.

2.2. Preparation of bacterial suspensions used for inoculation

The strains were reactivated from frozen (�80 �C) glycerolstocks in a broth medium, and cultures were then streaked on anagar medium and incubated in the conditions described in Table 1.Cell suspensions were prepared from bacterial colonies collectedfrom agar plates with a sterile loop (about 1 ml) and suspended in10 mL of a 9 g/L NaCl solution. Preliminary experiments were car-ried out to determine the optical density and viable counts of thesebacterial suspensions, so as to ensure inoculation with an accuratenumber of viable cells. The day of the experiment, the cell sus-pensions were used immediately after preparation to inoculate thecurd-based medium.

2.3. Preparation of curd-based medium

A curd-based medium was prepared from a fresh curd of semi-hard cheese, provided by an industrial cheesemaker. This cheesewas manufactured from pasteurized milk and inoculated only witha commercial lactococci starter, according to the usual cheesemanufacture process. Blocks (4 kg) of a non-brined curd (52e53%dry matter, 48e50% fat) were cut and stored wrapped in aluminumfoil in plastic bags under vacuum at �20 �C. The curd was left to

Growth conditions in brothc Growth conditions in agarc

T/Atmosphere Time, h T/Atmosphere Time, h

30 �C/agitation 24 30 �C/AE 7230 �C/agitation 24 30 �C/AE 4830 �C/agitation 24 30 �C/AE 7230 �C/agitation 24 30 �C/AE 7237 �C/AE 24 37 �C/AN 2437 �C/AE 24 37 �C/AN 4830 �C/AE 24 30 �C/AN 2430 �C/AE 24 30 �C/AN 2430 �C/AE 24 30 �C/AE 4830 �C/AE 24 30 �C/AN 120e14430 �C/AE 48 30 �C/AE 4830 �C/AE 48 30 �C/AE 48

oires Standa.st extract lactate broth; BHI-YE: brain heart infusion yeast extract, TSA trypticase soy

T. Poga�ci�c et al. / Food Microbiology 46 (2015) 145e153 147

thaw at 4 �C for 24 h before use. The exterior part (1 cm of eachside) of a 4 kg piece of curdwas discarded, and the interior part waspulverized using a Magimix blender (Magimix, Vincennes). Onepart curd was homogenized with two parts (weight/weight) ofsterilized (115oC/15 min) solution containing peptone (pancreaticdigest of meat type 2, Biokar Diagnostics Beauvais) 1.2 g/L, NaCl18 g/L, and lactose 1.2 g/L, using a laboratory Waring blender(Waring, Stamford) for 5 min (3 min at slow speed, followed by2 min at fast speed). Aliquots (10 mL) of the homogenized curd-based medium were transferred into 20 mL tubes, sterilized(110 �C for 15 min) and vortexed. The medium was stored at 4 �Cuntil use. Immediately before inoculation of the curd-based me-dium,12 mL of a filter-sterilized (0.2 mm-Sartorius, Aubagne) and 10-fold (v/v) diluted absolute ethanol (Sigma Aldrich, Saint-QuentinFallavier, France) solution were added to the curd medium to pro-mote the formation of the ethyl esters, as ethanol is considered thelimiting factor of ethyl ester synthesis in some cheeses.

2.4. Inoculation and incubation of curd-based medium

The curd-based medium was vortexed and inoculated with thebacterial suspensions, prepared as described above, to obtain2 � 106 colony-forming units (CFU)/mL. After inoculation the me-dium was vortexed and incubated in a water bath set to a thermalcycle monitored by a CINAC system (Alliance Instruments, France)chosen to mimic the first steps of semi-hard cheese manufacturingconditions: 32 �C for 95 min, ramped up to 38 �C over 15 min,maintained at 38 �C for 60 min, decreased from 38 �C to 36.5 �Cover 2 h, decreased from 36.5 �C to 15 �C over 20 h. Cultures werethen further incubated at 15 �C for 5 weeks. They were analyzed forbacterial counts using the media and conditions described inTable 1, pH values and volatiles at two incubation times: after 24 h(end of thermal cycle), and after incubation for 5 weeks. Controls(non-inoculated sterilized curd-based medium) were incubatedand analyzed under the same conditions. All experiments wereperformed in duplicate. Culture samples were stored at �80 �Cuntil analysis of volatile compounds.

2.5. Preparation of standards for GC-MS analysis

Standards of neutral volatile compounds were used i) to opti-mize the extraction step, ii) to generate standard curves, and iii) tocheck the response of the HS-trap GC-MS system during the runs ofsample analyses. Two solutions of standard compounds were pre-pared: one of neutral volatiles, and another of short-chain fattyacids.

Neutral standard compounds with various boiling points andchemical functions were chosen: four esters (ethyl acetate, ethylpropanoate, ethyl butanoate and ethyl hexanoate), two aldehydes(3-methylbutanal and benzaldehyde), one ketone (2-heptanone),2,3-butanedione, dimethyl disulfide and 3-methylbutanol. A firstsolution of compounds was prepared by combining approximately100 mg of each compound, except 3-methylbutanol (5000 mg). Analiquot of this mixture (7 mg) was diluted in 750 mg of methanol(SigmaeAldrich, �99.8%). This stock solution was kept frozenat �20 �C. The day of use, the stock solution was further diluted indeionized boiled water to prepare solutions of mixed standards atconcentrations ranging from 5 to 1200 ng/g each (from 260 to50,000 ng/g for 3-methylbutanol). For short-chain fatty acids, thestandard mix 46975-U (Supelco, SigmaeAldrich) containing acetic,propanoic, butanoic, 2-methylpropanoic, pentanoic, 3-methylbutanoic, hexanoic, 4-methylpentanoic and heptanoicacids at 10 mM each was used to prepare standard solutions from20 to 1000 ng/g each. For acetic and propanoic acids, standard so-lutions at concentrations up to 5000 mg/g were also prepared from

pure compounds (Sigma Aldrich, �99%) diluted in deionized boiledwater. The pH of these standard acid solutions was adjusted to6.35 ± 0.15 with NaOH.

2.6. Extraction of volatile compound using headspace trap

A Perkin Elmer Turbomatrix HS-40 trap automatic headspacesampler with trap enrichment was used to extract volatiles. Ali-quots (2.5 g) of cultures were placed in 22 mL PerkinElmer vialswith polytetrafluorethylene (PTFE)/silicone septa. The principle ofthe HS-trap method has been previously described in detail (Baraniet al., 2006; Schulz et al., 2007). It includes several steps: equili-bration, pressurization, trap load, trap dry-purge, trap desorptionand trap hold. The final conditions retained after optimizationwerethe following: Samples were warmed for 15 min at 65 �C (equili-bration). The vials were then pressurized for 1 min at 207 kPa withthe carrier gas (helium), by introducing a needle through theseptum. A Tenax™ trap at 35 �C was then loaded for 2.3 min byallowing the pressure to fall through the trap, permitting theadsorption and the concentration of the analytes of the headspaceon the trap. The trap load was repeated twice for each vial trap. Theadsorbed water was then removed by purging helium through thetrap (dry purge: 3 min). The trap was heated at 250 �C for 0.1 minand backflushed at 164 kPa, leading to desorption of the analytes,which were then transferred to the GC through a transfer linemaintained at 150 �C, with an injection time of 0.6 min. The trapwas held at 250 �C for 5 min.

The extractionwas optimized by varying four parameters, testedin a 24 full factorial experiment in duplicate: equilibration tem-perature (50 �C or 65 �C), equilibration time (15 min or 30 min),sample mass (2.5 or 5.0 g), and number of extraction cycles (1 or 2).Two traps were also compared: Tenax™ and “Air Monitoring Trap”(Perkin Elmer). This optimization was carried out for a 20 ng/gmixed standard solution and for samples of grated/homogenizedEmmental cheese in boiled water (1:3, w/w). For this step of opti-mization, one specific ion was used to quantify each compoundusing the Turbomass software.

The limits of detection (LOD), limits of quantification (LOQ) andlinearity ranges were determined for neutral and acid standardcompounds.

2.7. Analysis of volatile compounds using GC/MS

Volatiles were analyzed using a Clarus 680 gas chromatographcoupled to a Clarus 600T quadrupole mass spectrometer (PerkinElmer, Courtaboeuf, France). They were separated on an Elite_5MScapillary column (60 m � 0.25 mm � 1 mm; Perkin-Elmer), withhelium as the mobile phase. The initial temperature of the oven,35 �C, was maintained for 5 min. The increase of temperature wasperformed at a rate of 7 �C/min up to 140 �C and then at 13 �C/minup to 280 �C. The mass spectrometer was operated in the scanmode (scan time 0.3 s, interscan delay 0.03 s) within a mass rangeofm/z 29 to 206. Ionizationwas done by electronic impact at 70 eV.

All samples were analyzed in the same GC-MS run. Standardswere regularly injected to verify the absence of instrumental drift ofthe GC-MS system. Blank samples (boiled deionized water) werealso injected to check the absence of carry-over.

Volatile compounds were identified by comparison of massspectra and retention times with those of authentic standardspurchased from SigmaeAldrich (St. Quentin Fallavier, France), onthe basis of their retention index and mass spectral data from theNIST 2008 Mass Spectral Library (Scientific Instrument Services,Ringoes, NJ, USA). Some compounds were tentatively identified onthe basis of mass spectral data only when data on retention indiceswere not available.

Table 2Viable counts (colony-forming units) and pH after 24 h of cheese-related strainsincubated under conditions mimicking the thermal cycle during cheese manufac-ture (t1) and after 5 weeks further incubation at 15 �C (t2).

Strain Growth, CFU/mL pH

t1 t2 t1 t2

Brachybacterium articum LSBA53 2.9E þ 07 1.1E þ 07 5.36 5.39Brachybacterium tyrofermentans LSBT17 2.9E þ 07 1.4E þ 07 5.41 5.40Brevibacterium aurantiacum LSBA57 1.0E þ 07 2.7E þ 07 5.39 5.38Microbacterium gubbeenense LSMG39 3.7E þ 06 4.9E þ 08 5.35 5.45Lactobacillus fermentum LSLF202 2.5E þ 08 2.2E þ 07 5.05 5.19Lactobacillus helveticus CIRM-BIA108 1.6E þ 08 3.9E þ 08 5.08 4.89Lactobacillus paracasei LSLP248 9.3E þ 08 1.4E þ 08 4.91 5.00Lactobacillus sakei LSLS89 8.3E þ 07 1.7E þ 07 5.26 5.31Leuconostoc lactis CIRM-BIA1541 3.9E þ 07 1.3E þ 07 5.05 4.94Propionibacterium freudenreichii

CIRM-BIA14263.9E þ 06 1.7E þ 09 5.33 5.39

Hafnia alvei CIRM-BIA1620 4.9E þ 08 1.4E þ 09 5.41 5.56Hafnia alvei CIRM-BIA1621 2.6E þ 08 2.3E þ 08 5.39 5.44

T. Poga�ci�c et al. / Food Microbiology 46 (2015) 145e153148

2.8. Data processing

Data pre-processing was performed using PerkinElmer Turbo-mass software, version 5.4.2.1617. The GC-MS raw data files wereconverted to netCDF format with Data Bridge (Perkin Elmer, Wal-tham, Massachusetts, USA) for further analysis. GC-MS data wereprocessed by converting the raw data to time- and mass-alignedchromatographic peaks areas using the open source XCMS pack-age implemented with the R statistical language (Smith et al.,2006). The full width at half maximum was set to 5, the groupband-width to 3, and the other parameters were those by default.

2.9. Statistical and multivariate analysis

All statistical analyses were performed using software R http://www.r-project.org/.

A multivariate analysis of variance (MANOVA) was performedon the abundance of selected volatiles (11 standard compounds andeight selected volatiles for cheese samples) to determine whichfactors significantly affect the recovery of volatiles.

A principal component analysis (PCA) was performed onselected signals resulting from data processing using the XCMSpackage of R. PCA was performed on preprocessed, log10[x]-trans-formed and Pareto scaled data, using the package FactomineR of theR software.

An analysis of variance (ANOVA) was performed using R onselected signals to determine if strains significantly affected theabundance of each volatile. Means were compared using the leastsignificant difference (LSD) test.

3. Results and discussion

In fermented dairy products, starter as well as non-starterbacteria play a pivotal role in the development of flavor. In thepresent study we optimized a methodological approach to char-acterize the aromatic potential of very diverse cheese bacteria. Aworkflow was successfully developed. We first formulated a curd-based slurry medium and conditions of incubation to allowassessment of the growth and the aroma-producing potential of avariety of dairy bacteria under conditions relevant to cheese. Thenwe optimized the extractionmethod and analyzed volatiles. Finally,we applied a metabolomics-based method of data processing, fol-lowed by statistical and multivariate analysis.

3.1. Growth of varied bacterial species in the curd-based medium

The medium developed contained curd as the major ingredient,with added supplements. Lactose was added at a low concentrationto promote the growth of bacteria that do not use lactate or aminoacids as carbon source, but that are capable of growing during thefirst steps of cheese manufacture using milk lactose. Peptone wasadded to provide the peptides and amino acids that are released incheese during ripening. The preparation of a slurry medium insteadof a solid (agar) medium was chosen to facilitate the experimentsfor screening purposes.

The bacterial counts reached at the two incubation times aregiven in Table 2. All strains grew, withmaximal populations rangingfrom 7.4 to 9.2 log (CFU/mL). The highest populations wereobserved for Hafnia alvei CIRM-BIA1620 and Propionibacteriumfreudenreichii, which reached populations above 9.1 log CFU/mL.Ten out of the 12 strains grewmainly during the first phase of 24 h-incubation. The highest growth at 24 h was observed for Leuco-nostoc fermentum, Lactobacillus helveticus, Lactobacillus. paracasei,and H. alvei cultures, which exhibited more than 2 log increaseunder these conditions, reaching populations above 8.2 log (CFU/

mL), whereas other strains (most coryneform bacteria and theother LAB species) reached populations ranging from 7.0 to 7.9 logafter 24 h. Only Microbacterium gubbeenense and P. freudenreichiigrew slowly, reaching ~9 log (CFU/mL) after 5 weeks of incubationat 15 �C (Table 2). This was expected for both of these slow-growingspecies (Mounier et al., 2007; Thierry et al., 2011). For most strains,the population remained at similar levels after 5 weeks comparedto 24 h (below ± 0.5 log change). Three strains showed about 1 logdecrease during the 5 weeks incubation (L. fermentum, L. paracasei,and Lactobacillus sakei).

The growth of LAB generated, as expected, a significant decreaseof pH in the medium, initially at 5.3. The acidification was morepronounced for L. paracasei, L. helveticus, L. lactis and L. fermentum (-0.3 pH units after 24 h) compared to L. sakei (- 0.1 pH units)(Table 2). In all non-lactic cultures, the pH increased slightly(maximal changes of þ0.05 and þ0.2 pH units at 24 h and 5 weeks,respectively).

The bacterial species selected for this experiment (Table 1)belong to diverse groups of bacteria frequently used in dairyfermentation and isolated as non-starter bacteria from cheese orfrom natural whey culture (Brennan et al., 2001; Mounier et al.,2007; Poga�ci�c et al., 2013; Montel et al., 2014; Gatti et al., 2014).Several curd- (Martin et al., 2001), and cheese-based media (Goriet al., 2012; Sgarbi et al., 2013) have already been developed foraroma characterization of microorganisms but differ in composi-tion from our curd-based medium. The use of a medium that pro-vides only the nutrients available in cheese as the energy andnutrient sources is important, as previously mentioned (Sgarbiet al., 2013). All the bacteria tested grew under our medium con-ditions, most reaching maximal populations at 24 h incubation intemperature conditions simulating cheese manufacture. The addi-tion of a small amount of lactose in the curd medium likely favoredgrowth.

A crucial step that influences all subsequent results is theadjustment of concentration of viable bacterial cells for inoculationin the curd-based medium. In our experience, it is highly recom-mended that the suspensions of bacterial colonies, prior to use asinoculums, be checked for viability through appropriate pre-liminary experiments (optical density, CFU/mL) for every strain.This is a time consuming but indispensable experimental step. Inthe present study, for example, we observed that the standard platecounts of bacterial suspensions prepared from 72 h-agar platesfrom strain Brachybacterium tyrofermentans LSBT17 contained verylow numbers of cultivable cells. The same was observed for bothstrains of H. alvei on 48 h-agar plates. Therefore, we reduced to 24 h

Table 3Results of MANOVA: effect of four extraction parameters of the recovery of volatile compounds from a mixture of standard compounds and an Emmental cheese sample.

Factors Levels of factors Aqueous solution of 11 mixed standardsa Selected volatiles from cheeseb

P valuec Mean effect of factor (SD)d P valuec Mean effect of factor (SD)d

Thermostatization temperature (Temp) 65 vs 50 �C 2.33E-12 1.77 (0.29) 7.91E-10 2.39 (0.48)Cycle number (Cyc) 2 vs 1 2.86E-10 1.48 (0.09) 5.94E-09 2.03 (0.31)Sample amount (Amt) 5 vs 2.5 g 5.25E-09 1.15 (0.19) 1.45E-06 0.86 (0.11)Thermostatization time 30 vs 15 min NSe 0.93 (0.03) 0.031 1.21 (0.12)Temp:Cyc 5.06E-04 1.61E-04Temp:Amt 3.76E-04 0.012Cyc:Amt 0.072 5.87E-03

a A mixture of 11 volatiles with various boiling points and chemical functions (see Material & methods).b The abundance of 8 volatiles was quantified: diacetyl, 2-butanone, 2-methylbutanal, 2-pentanone, acetoin, 3-methylbutanol, 2-methylbutanol, 2-heptanone and 2-

nonanone.c Probability, from MANOVA function of R software.d Mean ratio was calculated as the mean of the ratio of abundance of each compound at each level.e NS: not statistically significant (P value > 0.05).

T. Poga�ci�c et al. / Food Microbiology 46 (2015) 145e153 149

the time of incubation of agar plates for some strains to maximizethe number of viable cells collected.

3.2. Optimization of extraction of volatile compounds by HS-trap-GC-MS

Four parameters of extraction were tested to optimize theextraction of volatiles using HS-trap, from two types of samples: anaqueous solution of mixed standard compounds, and cheese sam-ples. The results of the MANOVA are shown in Table 3. Two pa-rameters, the equilibration temperature and the number ofextraction cycles, significantly increased the recovery rates of allvolatiles, regardless of the sample type. The increase of tempera-ture from 50 �C to 65 �C induced a fold change ranging from 1.3 to3.0, whereas the use of 2 cycles of extraction instead of 1, a foldchange ranging from 1.4 to 2.4, depending on the volatile com-pound and the type of sample. The amount of sample placed in thevial of extraction had a significant, but smaller effect. The use of atwo-fold greater sample (5 rather than 2.5 g) increased the abun-dance of volatiles by 15%, on average, for standards, but decreasedthe abundance of volatiles by 14%, on average, for cheese samples.The increase of the equilibration time from 15 to 30 min did not

Table 4Limits of detection (LOD), limits of quantification (LOQ) and linearity ranges of theanalysis of aqueous solutions of mixed standard compounds analyzed by HeadspaceTrap GC-MS.

Compound Ion ofquantification

Linearityrange, ng/g

R2 LOD,ng/g

LOQ,ng/g

2,3-butanedione 86 0e1000 0.995 5 16Ethyl acetate 88 0e900 0.990 5 16Ethyl propanoate 102 0e1000 0.974 3 10Ethyl butanoate 116 0e900 0.924 3 9Ethyl hexanoate 99 0e160 0.933 5 173-methylbutanal 58 0e350 0.942 4 12Benzaldehyde 105 0e1000 0.967 4 122-heptanone 58 0e100 0.960 8 26Dimethyl disulfide 94 0e250 0.890 2 73-methylbutanol 70 0e4000 0.962 586 1758Acetic acid 60 0e500 0.886 520 1565Propanoic acid 74 0e600 0.817 450 13542-methyl propanoic acid 73 0e900 0.817 38 114Butanoic acid 60 0e800 0.853 57 1713-methylbutanoic acid 60 0e900 0.838 33 100Pentanoic acid 60 0e900 0.837 38 114Hexanoic acid 60 0e1100 0.844 109 3264-methylpentanoic acid 73 0e1000 0.817 83 250

have a significant effect for standards, but increased the abundanceof volatiles by 20%, on average, for cheese samples. Some significantinteractions were also observed between the studied parameters(Table 3).

Two types of traps were compared. The recovery was two-foldhigher on the Tenax™ trap, on average, compared to the “Airmonitoring” trap, regardless of the type of sample (data notshown).

Following this optimization step, the parameters of extractionwere set to maximize the recovery rate (equilibration time andtemperature: 15 min at 65 �C, 2 cycles of extraction, 2.5 g-samples,Tenax™ trap).

Under these conditions, LOD, LOQ, and linearity range wereestablished formixtures of neutral and acid standards (Table 4). Theextraction of volatiles was linear over a large range of concentra-tions varying from 100 to 1100 ng/g. The mean LOD and LOQ werearound 4 ng/g and 14 ng/g, respectively, for all neutral volatilesexcept 3-methylbutanol. For this alcohol and for all acids, the meanLOD and LOQ were 10e100 fold greater (50e500 ng/g and100e1700 ng/g, respectively).

HS-based extraction techniques have become popular for theextraction of food aroma compounds because they are relativelysimple, involve minimal sample preparation, and extract volatilesunder “mild” conditions. HS methods are divided into staticheadspace and dynamic headspacemethods. The former is a simplebut poorly sensitive method, whereas the latter, which involves apurge step and concentration of volatiles on a trap, is highly sen-sitive but cannot be automated for food samples (Jele�n et al., 2012).Among the diversity of automated HS-based methods of volatileextraction, HS-SPME is the most widely used method for foodproducts (Jele�n et al., 2012). The principle of HS-SPME is theadsorption of the HS volatiles on a coated silica fiber situated insidea needle. The small quantity of adsorbent, however, may limit theadsorption of some volatiles and induce competition between thevolatiles present, leading to variability of the profile of compoundsextracted. In this study, we chose a new patented (Tipler andMazza, 2004) HS-trap technique recently developed. In contrastto SPME, which applies small fibers with sorbent volumes of0.94 mm3, the HS-trap used in this study is packed with a signifi-cantly greater volume of sorbent, 160 mm3 (Schulz et al., 2007).

The data about the LOD/LOQ are scarce in the literature andmainly concern the analysis of fermented drinks, fruits and vege-tables by SPME (Jele�n et al., 2012). The LOD/LOQ values depend onthe nature of sample and of analyte, the equilibration conditions,and the type of adsorbent. Values ranging from ~0.01 ng/g to~50 ng/g have been reported. Our results are within this range forneutral compounds. Acids were extracted with lower

Table 5Identified aroma compounds in cultures in a curd medium of cheese-related bacteria strains incubated under conditions mimicking the thermal cycle during cheesemanufacture and followed by 5 weeks further incubation at 15 �C. ion (m/z) of quantification, and P-value of the ANOVA showing the effect of the culture on the amount of thecompound.

N RIa Compounds (other name or code) m/z Identificationb P valuec Ratio Ad Ratio Bd Max value in culturee

1 471 Ethanol 45 S, RI, DB 1.03E-11 5.1 26.5 LSLF2022 517 Dimethyl sulphide (DMS) 62 S, RI, DB 3.99E-12 6.0 35.6 CIRM16203 554 1-propanol 31 RI,DB 3.87E-11 31.6 33.7 CIRM16204 556 2-methylpropanal 41 S, RI, DB 5.56E-09 0.6 26.3 LSLS895 589 2,3-butanedione (diacetyl) 86 S, RI, DB 7.91E-13 13.0 48.3 LSLP2486 597 2-butanone 43 S, RI, DB 3.27E-07 3.6 3.5 CIRM14267 607 2-butanol 59 S, RI, DB 9.94E-09 58.5 48.6 LSLF2028 613 Ethyl acetate (ethyl C2) 70 S, RI, DB 5.38E-09 84.6 56.4 CIRM16209 622 Acetic acid (C2) 60 S, RI, DB 2.01E-03 15.0 12.6 CIRM142610 629 2-methylpropanol 74 S, RI, DB 5.81E-11 120.3 133.4 CIRM162011 659 3-methylbutanal 58 S, RI, DB 1.41E-03 5.8 850.1 CIRM162012 666 1-butanol 56 S, RI, DB 0.08 9.7 28.0 CIRM162013 666 2-methylbutanal 57 S, RI, DB 8.94E-09 3.3 88.9 CIRM162014 679 2-propanone-1-hydroxy 74 S, RI, DB 5.20E-10 11.6 29.8 LSLP24815 688 2-pentanone 86 S, RI, DB 0.02 1.0 1.8 CIRM142616 713 propanoic acid (C3) 74 S, RI, DB 6.57E-12 35.0 123.6 CIRM142617 695 2,3-pentanedione 100 S, RI, DB 0.02 8.5 60.9 LSMG3918 701 Pentanal 86 RI, DB 8.72E-05 0.9 2.2 CIRM154119 703 2-pentanol 45 S, RI, DB 3.78E-06 5.8 28.2 LSLF20220 711 Ethyl propanoate (ethyl C3) 102 S, RI, DB 2.59E-04 523.8 594.5 CIRM142621 712 2-butanone-3-hydroxy (acetoin) 45 S, RI, DB 1.62E-10 21.4 272.0 LSLS8922 739 3-methylbutanol 70 S, RI, DB 1.91E-10 3467.6 1405.4 CIRM162023 745 2-methylbutanol 56 S, RI, DB 1.75E-10 1144.3 747.4 CIRM162024 752 Dimethyl disulfide (DMDS) 94 S, RI, DB 2.37E-14 142.3 148.0 CIRM162125 756 3-methyl-2-pentanone 100 RI, DB 2.58E-11 481.4 565.2 LSBA5326 770 1-pentanol 41 S, RI, DB 3.24E-05 5.5 3.7 LSLF20227 781 Butanoic acid (C4) 60 S, RI, DB 1.04E-04 57.2 19.4 CIRM142628 792 2-hexanone 100 S, RI, DB 4.38E-03 1.3 2.0 CIRM142629 800 Ethyl butanoate (ethyl C4) 116 S, RI, DB 6.23E-09 164.6 240.2 LSBA5330 803 Hexanal 56 RI, DB 1.33E-03 0.7 12.3 LSLF20231 836 3-methylbutanoic acid (iC5) 87 S, RI, DB 6.16E-07 510.0 435.4 LSLP24832 845 2-methylbutanoic acid (aC5) 74 RI, DB 1.71E-05 86.3 100.0 LSLP24833 849 2-hexenal 98 RI, DB 2.11E-02 1.6 2.4 LSLP24834 871 1-hexanol 69 S, RI, DB 1.37E-07 16.3 7.5 CIRM154135 891 2-heptanone 114 S, RI, DB 6.16E-04 1.1 2.6 CIRM142636 895 Ethyl pentanoate (ethyl_C5) 103 RI, DB 3.57E-12 13.5 19.2 LSBA5337 900 2-heptanol 98 RI, DB 9.98E-10 113.6 201.6 LSLF20238 903 Heptanal 86 S, RI, DB 5.71E-07 1.4 18.1 CIRM10839 909 Methyl hexanoate (methyl C6) 99 RI, DB 5.35E-08 391.6 552.9 CIRM10840 952 Hexanoic acid (C6) 60 S, RI, DB 1.29E-03 87.8 129.7 LSBT1741 961 Benzaldehyde 106 S, RI, DB 9.30E-04 2.0 27.5 CIRM10842 969 Ethyl hexanoate (ethyl C6) 88 S, RI, DB 2.07E-28 1648.6 2140.3 LSBA5343 972 Dimethyl trisulfide (DMTS) 126 RI, DB 2.63E-12 90.0 43.1 CIRM162144 979 Octanal 84 S, RI, DB 1.70E-04 1.5 5.6 CIRM10845 1052 2-nonanone 58 S, RI, DB 0.03 1.3 3.6 CIRM154146 1078 2-nonanol 126 RI, DB 1.27E-10 30.8 43.7 LSLF20247 1104 Nonanal 82 S, RI, DB 8.19E-05 1.5 3.4 CIRM10848 1128 Phenylpropanone 134 DB 0.45 21.6 16.8 LSMG3949 1183 Ethyl octanoate (ethyl C8) 88 S, DB 1.52E-11 487.4 806.2 LSBA5350 1218 3-methylbutyl hexanoate (methylbutyl C6) 70 DB 1.12E-08 313.6 471.7 LSBA5351 1282 2-undecanone 71 DB 0.18 1.4 3.2 LSBT1752 1390 Ethyl decanoate (ethyl C10) 88 DB 7.14E-07 35.0 97.3 LSBA53

a RI, Kovats retention index.b Compounds identified on the basis of: S, retention time and mass spectrum from standard; RI, retention index; DB, mass spectral data Library NIST.c P-value of ANOVA.d Ratio A: Maximal ratio of abundance between cultures and control medium; ratio B: Maximal ratio of abundance between cultures.e CIRM for CIRM-BIA.

T. Poga�ci�c et al. / Food Microbiology 46 (2015) 145e153150

reproducibility and sensitivity. They are important flavor com-pounds in cheese but are not extracted by all HS methods. Ourresults show that HS trap extraction allows the automated extrac-tion of both neutral and acidic volatiles with satisfactory sensitivityand reproducibility.

3.3. Production of volatile compounds

A metabolomic approach was used to facilitate the handling ofthe large data sets generated from GC-MS analyses. MS metab-olomics has emerged over the past decade in many scientific areas,

including food science, leading to the subsequent development ofworkflows and several dedicated software programs (Antignacet al., 2011) to perform automatic peak detection and alignmentof peak retention times. We chose the package XCMS of R (Smithet al., 2006) because it is an open-source tool widely applied toprocess data from chromatography-MS-based metabolomic ap-proaches. The extraction of peak information from the data set isimportant to guarantee reliable statistical analyses and biologicalinterpretation (Antignac et al., 2011). The package XCMS wasinitially developed for liquid chromatography (LC)-MS data, but theparameters of the software can be adapted for GC-MS data. In this

Fig. 1. Results of Principal Component Analysis on the abundance of volatiles in cultures of cheese-related bacteria showing the first two principal components. Plots of the first twocomponents. a, score plot: cultures are abbreviated as the first letter of genus and species, see Table 1 for full names, letters a and b in the name represent each of the two biologicalreplicate cultures. b, loading plot (volatiles, see Table 5). The names of the volatiles poorly represented on this plot are not shown.

T. Poga�ci�c et al. / Food Microbiology 46 (2015) 145e153 151

context, XCMS has been mainly applied to process data of analysisof solutes after derivatization. Recently, XCMS has also been suc-cessfully used to analyze the results of GC-MS data in the field ofcheese aroma formation (Le Boucher et al., 2013).

In the present study, data processing using XCMS generated2863 signals (i.e. onem/z at one time). From these signals, a total 65volatiles could be identified, since many ions (m/z) are generatedand detected from a single compound. Alkanes (hexane, heptane,among others) and contaminants (such as trichloromethane) werenot retained in the data set because they are not flavor-related, and

some compounds detected in the curd and known to originate frommilk (such as limonene) were also excluded.

Fifty-two aroma compounds were identified, including nineesters, six acids, 12 ketones, 12 alcohols, 10 aldehydes, and threesulfur compounds (Table 5). Interestingly, compounds resultingfrom various origins were produced. For example, acetic and pro-pionic acids result from the fermentation of lactic acid byP. freudenreichii, and diacetyl from citrate or lactose conversion inlactic cultures. Branched-chain (BC) and sulfur compounds resultfrom the catabolism of valine, leucine, isoleucine, and methionine

T. Poga�ci�c et al. / Food Microbiology 46 (2015) 145e153152

(Yvon and Rijnen, 2001; Martínez-Cuesta et al., 2013). Free fattyacids result from milk fat lipolysis (Collins et al., 2003). Sinceethanol is considered the limiting factor of ethyl ester synthesis insome cheeses, such as cheddar cheese (Liu et al., 2004; Urbach,1993) and Swiss cheese (Richoux et al., 2008), ethanol was addedto the curd medium at a concentration of 92 mg/g to promote theformation of ethyl esters. Seven ethyl esters were detected: ethylacetate, propanoate, butanoate, pentanoate, hexanoate, octanoateand decanoate.

All volatiles were detected at both incubation times (data notshown). The abundance of most of them increased between t1 andt2 in most cultures, by a mean factor of 2, with high differencesdepending on the volatile and the strain. For example, ethyl prop-anoate varied by a factor above 100 between t1 and t2 in someP. freudenreichii cultures, as did dimethyl disulfide, 3-methyl-2-pentanone and 2-heptanol in some cultures of actinobacteria.Only some aldehydes, such as hexanal and benzaldehyde,decreased in concentration over time in most cultures (data notshown). Therefore only results at t2 are presented.

The results of the ANOVA performed on the abundance of eachvolatile compound at t2 showed significant differences betweencultures for most compounds (49 out of 52, Table 5). However, theabundance of volatiles significantly differed from the control me-dium in an only small number of cultures, depending on thecompound considered. The fold-change between the aroma-producing cultures and the controls reached very high values.Maximum fold-changes were as high as 20 for carbonyl com-pounds, 80 for FFA frommilk fat lipolysis, 1000 for ethyl esters, and3000 for BC compounds.

A PCA (Fig. 1) was performed to summarize the differences involatile profiles between the strains studied. The first two PCsaccounted for 50% of the total variability. All the inoculated cultureswere distinct from the control samples, except that of L. helveticus.The biological replicates of cultures appeared closely localized,demonstrating good global reproducibility of the experiments andanalyses.

PC1, describing 31.5% of the variability, differentiated the twoBrachybacterium strains, Brachybacterium articum andB. tyrofermentans, from the other cultures and controls, on the basisof the concentration of acids (butanoic and hexanoic acids) andmany esters (ethyl butanoate, ethyl pentanoate, ethyl hexanoate,ethyl octanoate, ethyl decanoate and 3-methylbutyl hexanoate). Forexample, the amounts of ethyl butanoate, 3-methylbutyl hex-anoate, and ethyl hexanoate were more than 100-fold higher in thecultures of Brachybacterium strains compared to the control me-dium and most other cultures (ratio A and B in Table 5).

PC2, describing 18.8% of the variability, differentiated cultureson the basis of sulfur compound abundance, positively associatedwith PC2, and aldehydes, negatively associated with PC2. The threesulfur compounds identified (DMS, DMDS, and DMTS) were asso-ciated with both strains of H. alvei. In the culture of H. alvei CIRM-BIA1620, for example, the amounts of DMS, DMDS, and DMTSwere significantly higher (6, 142, and 90-fold higher, respectively)compared to the controls. Straight-chain aldehydes (pentanal,hexanal, heptanal, octanal, nonanal) and benzaldehyde wereassociated with controls and with cultures of L. fermentum andL. helveticus. For example, the abundance of hexanal was 13-foldhigher in the controls compared to the cultures of H. alvei andL. paracasei. BC-acids and BC-alcohols were also positively associ-ated with PC1 and PC2. These BC-compounds were present insignificantly higher amounts in cultures of H. alvei CIRM-BIA1620,the two Brachybacterium strains and the P. freudenreichii strainthan in controls. For example, the amounts of 2-methylpropanol, 3-methylbutanol, and 2-methylbutanol were 120, 3467 and 1144-foldgreater, respectively, in the culture of H. alvei CIRM-BIA1620

compared to the control medium (Table 5). The L. sakei strain wasassociated with acetoin, whereas L. paracasei was associated withdiacetyl and 2-propanone-1-hydroxy, detected in ~10e20-foldhigher concentrations in these cultures compared to the controland most other cultures (Fig. 1 and Table 5). PC3 and PC4 accountedfor 13.6% and 8.9% of the total variance, respectively (data notshown). PC3 clearly separated the P. freudenreichii strain from othercultures, on the basis of its production of a range of compounds,including acids (acetic, propanoic and butanoic acids), ethyl prop-anoate, and ketones (2-butanone, 2-hexanone and 2,3-pentanedione). PC4 separated the L. fermentum strain because ofits higher production of ethanol, 1-hexanol, and secondary alcohols(2-butanol, 2-pentanol, 2-heptanol, and 2-nonanol). These resultsare in agreement with previous reports on the ability of cheese-related bacteria to produce aroma compounds (Deetae et al.,2007; Thierry et al., 2011; Irlinger et al., 2012; Sgarbi et al., 2013).Volatile profiling, followed bymultivariate analysis has beenwidelyapplied to analyze the aroma potential of different cheese micro-organisms. For example, the volatiles profiles were studied for non-starter LAB strains (Sgarbi et al., 2013), mixed LAB starter cultures(De Bok et al., 2011), surface bacteria (Deetae et al., 2007), Gramnegative bacteria (Irlinger et al., 2012), or yeasts (Chen et al., 2012).However, despite the development of metabolomics, and inparticular metabolite profiling by GC-MS (Fiehn, 2008),metabolomics-based workflows have only rarely been used toprocess GC-MS data of volatile aroma compounds. Recently,another fingerprinting approach was proposed for the high-throughput screening of LAB, which combined the analysis of vol-atiles by static headspace -ultrafast GC/MS and their preprocessingby MetAlign, another software program for the pre-processing andcomparison of LC and GCeMS data (De Bok et al., 2011).

4. Conclusion

Foodmetabolomics-based studies are becomingmore intensive,aiming for deeper understanding of the function of food microbiota(Turnbaugh and Gordon, 2008; Le Boucher et al., 2013; Sgarbi et al.,2013). The methodological approach presented in this study can beapplied to evaluate very different genera of cheese-associatedbacteria for their ability to produce aroma volatile compounds,and particularly for their potential to enhance the flavor of semi-hard cheeses. The HS-trap/GC-MS method used was demon-strated to be efficient tool for volatile extraction. Themetabolomics-based workflow of data processing using XCMSfollowed by multivariate analyses facilitates data analysis. Thisapproach can be applied to screen the potential of non-starterstrains isolated from traditional fermented dairy products or fromdifferent origins, mixed cultures of strains.

Acknowledgment

This work was partly supported by a grant from the region ofBrittany, France (Project PROPAROM e AAP CRITT CBB 2012).Tomislav Poga�ci�c is grateful to UMR1253 (INRA, Agrocampus Ouest)for providing a 7 month (MarcheSeptember 2013) postdocresearch and training period.

References

Aberl, A., Coelhan, M., 2012. Determination of volatile compounds in different hopvarieties by headspace-trap GC/MSein comparison with conventional hopessential oil analysis. J. Agric. Food Chem. 60, 2785e2792.

Antignac, J., Courant, F., Pinel, G., Bichon, E., Monteau, F., Elliott, C., Bizec, B., 2011.Mass spectrometry-based metabolomics applied to the chemical safety of food.Trends Anal. Chem. 30, 292e301.

T. Poga�ci�c et al. / Food Microbiology 46 (2015) 145e153 153

Barani, F., Dell'Amico, N., Griffone, L., Santoro, M., Tarabella, C., 2006. Determinationof volatile organic compounds by headspace trap. J. Chromatogr. Sci. 44,625e630.

Brennan, N.M., Brown, R., George, K., Building, I.V., Goodfellow, M., Ward, A.C.,Beresford, T.P., Vancanneyt, M., Cogan, T.M., Fox, P.F., 2001. Microbacteriumgubbeenense sp. nov., from the surface of a smear-ripened cheese. Int. J. Syst.Evol. Microbiol. 51, 1969e1976.

Chen, L., Cui, J., Ding, Q., Ma, Y., Chen, L., Dong, J., Jiang, T., Maubois, J., 2012. Theeffect of yeast species from raw milk in China on proteolysis and aroma com-pound formation in Camembert-type cheese. Food Bioproc. Technol. 5,2548e2556.

Collins, Y.F., McSweeney, P.L.H., Wilkinson, M.G., 2003. Lipolysis and free fatty acidcatabolism in cheese: a review of current knowledge. Int. Dairy J. 13, 841e866.

Croissant, A.E., Watson, D.M., Drake, M.A., 2011. Application of sensory andinstrumental volatile analyses to dairy products. Annu. Rev. Food Sci. Technol. 2,395e421.

Curioni, P.M.G., Bosset, J.O., 2002. Key odorants in various cheese types as deter-mined by gas chromatography-olfactometry. Int. Dairy J. 12, 959e984.

De Bok, F., Janssen, P.W.M., Bayjanov, J.R., Sieuwerts, S., Lommen, A., Van HylckamaVlieg, J.E.T., Molenaar, D., 2011. Volatile compound fingerprinting of mixed-culture fermentations. Appl. Environ. Microbiol. 77, 6233e6239.

Deetae, P., Bonnarme, P., Spinnler, H.E., Helinck, S., 2007. Production of volatilearoma compounds by bacterial strains isolated from different surface-ripenedFrench cheeses. Appl. Microbiol. Biotechnol. 76, 1161e1171.

Dhaisne, A., Guellerin, M., Laroute, V., Laguerre, S., Cocaign-Bousquet, M., LeBourgeois, P., Loubiere, P., 2013. Genotypic and phenotypic analysis of dairyLactococcus lactis biodiversity in milk: volatile organic compounds as discrim-inating markers. Appl. Environ. Microbiol. 79, 4643e4652.

Fiehn, O., 2008. Extending the breadth of metabolite profiling by gas chromatog-raphy coupled to mass spectrometry. Trends Anal. Chem. 27, 261e269.

Gatti, M., Bottari, B., Lazzi, C., Neviani, E., 2014. Invited review: microbial evolutionin raw milk long ripened cheeses produced using undefined natural wheystarters. J. Dairy Sci. 97, 573e591.

Gori, K., Sørensen, L.M., Petersen, M.A., Jespersen, L., Arneborg, N., 2012. Debar-yomyces hansenii strains differ in their production of flavor compounds in acheese-surface model. Microbiol. Open 1, 161e168.

Hassan, F.A.M., Gawad, M.A.M.A.E., Enab, A.K., 2013. Flavour compounds in cheese(review). Res. Precis. Instrum. Mach. 2, 15e29.

Irlinger, F., Yung, S.A.Y.I., Sarthou, A.S., Delbes-Paus, C., Montel, M.C., Coton, E.,Coton, M., Helinck, S., 2012. Ecological and aromatic impact of two Gram-negative bacteria (Psychrobacter celer and Hafnia alvei) inoculated as part ofthe whole microbial community of an experimental smear soft cheese. Int. J.Food Microbiol. 153, 332e338.

Jele�n, H.H., Majcher, M., Dziadas, M., 2012. Microextraction techniques in theanalysis of food flavor compounds: a review. Anal. Chim. Acta 738, 13e26.

Le Boucher, C., Courant, F., Jeanson, S., Chereau, S., Maillard, M.-B., Royer, A.-L.,Thierry, A., Dervilly-Pinel, G., Le Bizec, B., Lortal, S., 2013. First mass spec-trometry metabolic fingerprinting of bacterial metabolism in a model cheese.Food Chem. 141, 1032e1040.

Liu, S.Q., Holland, R., Crow, V.L., 2004. Esters and their biosynthesis in fermenteddairy products: a review. Int. Dairy J. 14, 923e945.

Maarse, H., Visscher, C.A., Willemsens, L.C., Nijssen, L.M., Boelens, M.H., 1994. Vol-atile Compounds in Food: Qualitative and Quantitative Data. Dairy Products.TNO Nutrition and Food Research, Zeist, The Netherlands, pp. 32e136.

Mariaca, R., Bosset, J.O., 1997. Instrumental analysis of volatile (flavour) compoundsin milk and dairy products. Le Lait 77, 13e40.

Marsili, R., 2011. Flavors and off-flavors in dairy foods. In: Fuquay, J.W., Fox, P.F.,McSweeney, P. (Eds.), Encyclopedia of Dairy Science, pp. 533e551.

Martin, N., Berger, C., Le Du, C., Spinnler, H.E., 2001. Aroma compound production incheese curd by coculturing with selected yeast and bacteria. J. Dairy Sci. 84,2125e2135.

Martínez-Cuesta, M.D.C., Pel�aez, C., Requena, T., 2013. Methionine metabolism:major pathways and enzymes involved and strategies for control and diversi-fication of volatile sulfur compounds in cheese. Crit. Rev. Food Sci. Nutr. 53,366e385.

Montel, M., Buchin, S., Mallet, A., Delbes-Paus, C., Vuitton, D.A., Desmasures, N.,Berthier, F., 2014. Traditional cheeses: rich and diverse microbiota with asso-ciated benefits. Int. J. Food Microbiol. 177, 136e154.

Mounier, J., Rea, C., O’Connor, M.O., Fitzgerald, G.F., Cogan, T.M., 2007. Growthcharacteristics of Brevibacterium, Corynebacterium, Microbacterium, and Staph-ylococcus spp. isolated from surface-ripened cheese. Appl. Environ. Microbiol.73, 7732e7739.

Neviani, E., Bottari, B., Lazzi, C., Gatti, M., 2013. New developments in the study ofthe microbiota of raw-milk, long-ripened cheeses by molecular methods: thecase of Grana Padano and Parmigiano Reggiano. Front. Microbiol. 4, 1e14.

Niimi, J., Eddy, A.I., Overington, A.R., Heenan, S.P., Silcock, P., Bremer, P.J.,Delahunty, C.M., 2014. Aromaetaste interactions between a model cheesearoma and five basic tastes in solution. Food Qual. Prefer. 31, 1e9.

Ordiales, E., Martín, A., Benito, M.J., Hern�andez, A., Ruiz-Moyano, S.,C�ordoba, M.D.G., 2013. Role of the microbial population on the flavor of thesoft-bodied cheese Torta del Casar. J. Dairy Sci. 96, 5477e5486.

Pedersen, T.B., Ristagno, D., McSweeney, P.L.H., Vogensen, F.K., Ard€o, Y., 2013. Po-tential impact on cheese flavour of heterofermentative bacteria from startercultures. Int. Dairy J. 33, 112e119.

Poga�ci�c, T., Mancini, A., Santarelli, M., Bottari, B., Lazzi, C., Neviani, E., Gatti, M., 2013.Diversity and dynamic of lactic acid bacteria strains during aging of a longripened hard cheese produced from raw milk and undefined natural starter.Food Microbiol. 36, 207e215.

Richoux, R., Maillard, M.B., Kerjean, J.R., Lortal, S., Thierry, A., 2008. Enhancement ofethyl ester and flavour formation in Swiss cheese by ethanol addition. Int. DairyJ. 18, 1140e1145.

Schulz, K., Dressler, J., Sohnius, E.-M., Lachenmeier, D.W., 2007. Determination ofvolatile constituents in spirits using headspace trap technology. J. Chromatogr.1145, 204e209.

Sgarbi, E., Lazzi, C., Tabanelli, G., Gatti, M., Neviani, E., Gardini, F., 2013. Nonstarterlactic acid bacteria volatilomes produced using cheese components. J. Dairy Sci.96, 4223e4234.

Smith, C.A., Want, E.J., O'Maille, G., Abagyan, R., Siuzdak, G., 2006. XCMS: processingmass spectrometry data for metabolite profiling using nonlinear peak align-ment, matching, and identification. Anal. Chem. 78, 779e787.

Steele, J., Broadbent, J., Kok, J., 2013. Perspectives on the contribution of lactic acidbacteria to cheese flavor development. Curr. Opin. Biotechnol. 24, 135e141.

Thierry, A., Deutsch, S.M., Falentin, H., Dalmasso, M., Cousin, F., Jan, G., 2011. Newinsights into physiology and metabolism of Propionibacterium freudenreichii. Int.J. Food Microbiol. 149, 18e27.

Tipler, A., Mazza, C., 2004. (PERKIN ELMER LAS, INC.), System and Method forExtracting Headspace Vapor. International Application Number PCT/US2004/011487.

Turnbaugh, J.P., Gordon, J., 2008. An invitation to marriage of metagenomics andmetabolomics. Cell 134, 708e713.

Urbach, G., 1993. Relations between cheese flavour and chemical composition. Int.Dairy J. 3, 389e422.

Wishart, D.S., 2008. Metabolomics: applications to food science and nutritionresearch. Trends Food Sci. Technol. 19, 482e493.

Yvon, M., Rijnen, L., 2001. Cheese flavour formation by amino acid catabolism. Int.Dairy J. 11, 185e201.

Zareba, D., Ziarno, M., �Scibisz, I., Gawron, J., 2014. The importance of volatilecompound profile in the assessment of fermentation conducted by Lactobacilluscasei DN-114 001. Int. Dairy J. 35, 11e14.