Electronic nose: Smelling the microbiological quality of cheese

Transcription

1 Electronic nose: Smellin te microbioloical quality of ceese Jeoros Triaas BiC-DTU, Tec. Univ. of Denmark Lynby, Denmark Tatjana van den Tempel Ceese Culture Tecnoloy, Cr. Hansen A/S. 97 Hørsolm, Denmark Per Væemose Nielsen BiC-DTU, Tec. Univ. of Denmark Lynby, Denmark Abstract Te volatile compounds produced by microoranisms involved in ceese production often vary in a very caracteristic way for a species or a roup of species. Measurements were made on te eadspaces of Danis blue ceese, Camembert ceese and ceese models wit electronic noses. An explorative data analysis on te sinals derived from te cemical sensors was performed in order to define te best profile parameters for estimatin te microbioloical quality. Keywords E-nose, Danis blue ceese, Camembert ceese, ceese models and multivariate analyses INTRODUCTION Te quality control of intermediate and final food products especially wit reard to undesirable components is of reat importance for te food industry. It is important tat te consumer as confidence tat tere are no microbial contaminants, toxins, off flavors and oter odors. Te development of innovative rapid detection systems wit potential for early detection is terefore required by te food industry. Te aim of tis study was to investiate te possibility of implementin electronic nose tecnoloy in order to evaluate te microbioloical quality of mould ceeses. In te experiments performed, electronic nose tecnoloy as been applied directly on representative Danis blue ceese samples, wic were contaminated by G. candidum. Additionally Camembert ceeses made wit different strains of P.camemberti were evaluated as well as ceese models simulatin fres ceese curd for Danis blue ceese inoculated wit pure mould and yeast cultures. Tis was done in order to et an objective caracteristic of te aroma profile concernin different contaminants and different strains used in ceese manufacturin. MATERIALS AND METHODS Danis blue as well as Camembert ceeses were collected from major manufacturers in Denmark. Danis blue ceeses were measured weeks after brinin. Contamination at tat ripenin stae was identified in reat numbers ( 7 ). Camembert ceeses were manufactured wit different commercial isolates of te mould P.camemberti and were analyzed on 6 consecutively weeks. Model ceeses were made wit of tese strains and were similar in terms of pysioloical, sensorial and e-nose profiles. Ceeses produced wit oter strains, were evaluated for teir maturity based on teir e-nose aroma profile models. Ceese models (ceese aar medium in 9 cm Petri dises) were manufactured from fres Danis blue ceese taken before brinin (inactivation of P.roqueforti was acieved by meltin te ceese mass). Tese models were inoculated wit, ml spore suspension ( spores/ml) of cultures from te ouse collection as well as oters (see table ). All models were placed at o C before profilin. Table. Microoranisms used for te ceese model experiments IBT number Microoranism P. roqueforti 9 P. caseifulvum different strains P. camemberti P. commune 9 G. candidum From KVL collection D. ansenii From KVL collection C. colicullosa An e-nose (model BH-: Bloodound Sensors Ltd. Leeds, UK), wic employed conductin polymer (polyaniline) sensors, was used in te experiments wit Danis blue ceeses and ceese models. Samples were taken from te eadspace of ceese slices. Slices were put in lass Petri dises in self-sealin plastic bas (Xcm). Bas were filled wit dry air (umidity<.%), secured and equilibrated at o C for. All experiments were performed in triplicate. Institute for Biotecnoloy, Tecnical University of Denmark Te Royal Veterinary and Aricultural University of Denmark

2 Te model afox-: Alpa M.O.S. (Multi Oranoleptic Systems), France, employs sensors and includes 6 Metal Oxide Semiconductor sensors (MOS), Conductive Polymer (CP) sensors and Quartz Crystal Microbalance (QCM) sensors. Tis instrument was used in te experiments wit te Camembert ceeses. All samples were cm cylindrical cuts from sample. Cuts were sealed in ml eadspace vial (6x. mm) and were put in an HS- autosampler (CTC Analytics AG, Switzerland). Te samples were incubated at o C for min. A µl from te eadspace was injected into te sensor cambers. Dry filtered airflow trou te sensors was adjusted to ml min -. Principal Component Analysis (A), Discriminant Partial Least Square (PLS-DISCR), Partial Least Squares Reression (PLS-R) and Soft Independent Modelin of Class Analoy (SIMCA) metods were performed usin te software packae Unscrambler (CAMO) version 7.6 SR. RESULTS AND DISCUSSION Danis blue ceeses Four caracteristic parameters of te response curves from te BH- were exported from te control software (see fiure ). Tese caracteristics were absorption (B: maximum rate of cane of resistance), desorption (C: maximum neative rate of cane of resistance), diverence (A: maximum step response) and area (E: area under te actual sensor curve). - - Scor Normal - - /D/D 8/D /D 9/D /D/D /D /D /D Correlation L di (X) /A /A /Ab 7/Ab b /Ab 8/Ab /Ab /A 6/Ab 9/Ab 7/Di /A b b 6/Di /Ab 7/Ar /Di /D /Ar /Di /Ar i /Ar /Di /Di /Ar /D 8/Di 9/Di /Ar /D /Ar i9/ar 8/Ar i /Ar /A /D i Fiure. A score and loadins plot from ood and contaminated ceeses. All variables used Te new A model could describe 9% of te total variation witin s. Te first accounted for 8% of te total variance and describes te differences between ood and contaminated ceeses wile accounts for 6% of te total variance and describes aain te differences between production days. Two classes of samples were cosen and used for SIMCA analysis. Tese models ( normal and infected ) were able to classify unknown samples very successfully (see fiure ).. Sample to Model Distance. Fiure. Response curves from e-nose and parameters used for multivariate analysis analysis carried out wit all variables (see fiure ) found tat s accounted for 8 % of te experimental variance. (6% of te total explained variance) was dominated by te difference between te ood and contaminated samples. (9%) is dominated by te variation between batces of ceeses and tis is due to te dayto-day variations of te ceeses, wic can be very i. Selection of sensor features wit elp of A elped improve te discrimination between ceeses. In a new A model only sensors were used.. Normal model Leverae 6 7 Fiure. SIMCA classification of unknown samples at % sinificance level (-week-old ceeses). Te discrimination power between tese models indicated te individual sensor variables tat discriminate best between te two ceese classes. A final analysis based on tese tree variables ave te best discrimination model were accounted for te differences between normal and infected ceeses wit a 97% of te total variance (see fiure ). Te sec-

3 ond (%) was not accounted for any differences between te normal and infected samples. PLS-Reression analysis based on tese variables (Y) and 8 sensorial caracteristics (X) tat describe te normal and te infected ceeses in terms of taste and smell yielded i correlation coefficients. in te relationsip of te sensors wit te compounds in te eadspace (see fiure ). In tis way te canes in sensor sinals associate to aroma compound canes in te eadspace.. Correlation Loadins (X and Y) Disulfide, dime -Nonanol -Heptanol.6. -Pentanol Normal Fiure. A score plot from normal and infected ceeses. Data from sensors Microbioloical findins were positively correlated to te e-nose variables, describin very well te rowt of te normal funi and te existence of oter funi in te ceeses. Normal ceeses ad ier ph tan te infected. Tis difference can be attributed to te differences in mould rowt []. Te main differences between te two ceese classes in terms of aroma detected in teir eadspace can be seen in table. Table. Total area % of compounds from Danis blue ceese eadspace Compound Normal Acetone Butanone.. -Propanol, -metyl Pentanone..9 Metyl Butanoate.. -Pentanol..6 -Butanol, -metyl-..9 -Hexanone.87. -Heptanone Heptanol.8. -Octanone.7.6 metylbutyl Butanoate.. -Nonanone.. -Undecanone..6 PLS-R analysis based on te e-nose variables (Y) and te aroma compounds (X) sowed ily positive correlation coefficients for many of te compounds measured. Reverse analysis (e-nose X variables and aroma compounds Y variables) ave valuable information concernin te relationsip of te sensors wit te.. -Butanol, -me -Nonanone -Undecanone -Propanol, -m -Octanone -metbu-butano -Pentanone -Heptanone Acetone Metylbutanoate -Butanone -Hexanone /Area /Div. /Area Fiure. Loadin plot from PLS-R analysis. (X = lo aroma compounds, Y = e-nose variables). Camembert ceeses Te response curves of te different sensors from tis experiment were used for analysis. Data used for te models was from to days old ceeses. analysis carried out wit all sensors (see fiure 6) found tat s accounted for 78% of te experimental variance. days days days days Fiure 6. A score plot of model ceeses. Wole response sinals used. (67% of te total explained variance) was dominated by te time canes occurrin to smell wit maturation. (%) is dominated by te variation of te different ceeses between and days. Tese s accounted for 8 % of te experimental variance. More s were not accounted for any caracteristics of te ceeses canes. Wit te elp of PLS-DISCR analysis a model based on te time canes (discrete variables) was made. Selection of variables sinificant on a % level was made. Based on tese variables a new simpler A model was introduced increasin te total explained variance. SIMCA analysis was performed based on tese last models.

4 Similar ceeses made wit different isolates to te model ceeses were identified. Te area of interest in tis particular type of ceeses is between te ae of and days. After days te ceeses were matured and ready for consumption avin all te caracteristics of a Camembert ceese []. Discrimination power model of te model of days on te model of days for tese ceeses elped to coose te important variables, wic were exported in a new reduced matrix for furter analysis. Tis model was improved slitly and described te maturity canes usin less tan % of te oriinal variables. Discrimination between and days ceeses was improved in tis model. Based on tese results a prediction model of ceese maturity (Y) was created based on e-nose responses (X). Results from tis analysis sowed tat te model was very ood in predictin te ae of oter strains, already juded as similar to te ones used for te model (see fiure 8). Te prediction performance was better wen te ceeses were reacin teir ready-to-eat ae ( days) Predicted Y days '_A_' '7_A_6' '_A_' '_A_' '_A_' '7 ' Samples Growt of microoranisms All samples were controlled for visible rowt for a period of 8 days. Yeasts were visible prior to funi (-6 days on te % model at o C). Funi sowed visible rowt only after 7-8 days (on te % model at o C). Minor contaminations of some plates were observed only after days at o C wit P. roquefortii. analysis was performed wit data from te BH after and days of incubation at o C. % salt model Te days model explained 7% of te total variation on s (see fiure 8). Te first (6%) described te existence of microoranisms or not I I I I I Fiure 8. A score plot from % plates ( days old). Te days model (see fiure 9) explained 7 % (: 6%, : 7%) of te total variation. Te variation described by is attributed to te various funi used as cultures. Blank plates tat were contaminated were clearly placed near to te inoculated plates. Fiure 7. Prediction of days old unknown Camembert ceeses by te e-nose responses. I Te ceese models Models ( and % NaCl) were made from fres Danis blue ceese. Averae values for a w and ph for tis type media are: Water activity:.99 ph:.7 Addition of salt (% NaCl) made te el unstable and very soft. Neiter te increase of aar from,% to % in te media nor te addition of lycerol resulted to a better el. Anoter serious problem of tis type media was tat tey ave i percentae of fat (~% of total weit) and bi surface terefore exposed to oxidation. For tis reason tey are not stable over time. Differences measured by e- nose tou, were only detected tree weeks from fabrication day I I I Fiure 9. A score plot from % plates ( days old). % salt model Results were similar to te ones wit te % model wit te exception of te samples inoculated wit G.candidum. Tese samples were more like te blanks and tis can be I

5 explained by te i sensitivity of tat particular funus to salt []. Te model % salt after days (see fiure 9) explains 7% of te variation on principal components (). 6% describes aain te main difference between blanks and inoculated plates wile (9%) is caracteristic for te funi G.candidum I I I G:candidum Fiure 9. A score plot days % salt model at o C after inoculation Identification of contaminants Based on A analysis te e-nose was able to differentiatiate between funi (see fiure ) and yeasts mainly on te % media after 8 incubation at o C. Media wit % salt were better for te discrimination of te salt tolerant moulds like P.camemberti and P.caseifulvum Fiure. A score plot days % salt model at o C after inoculation CONCLUSIONS Te BH E-nose system was successfully used in detectin te oriin and te microbioloical quality of Danis blue ceeses. Ceeses wit contamination were identified by e-nose wit te use of only few variables. Results were positively correlated to oter analyses (GC-MS, microbioloical and cemical). Selection of tese variables was possible by usin multivariate analyses metods as A, SIMCA and PLS-R. Te afox- model provided te means to model te smell of Camembert ceese made by specific microbial strains. Te application of multivariate analyses metods I I as A, SIMCA and PLS-R elped to locate te best features from te response sinals. Simplification of te model increased its effectiveness and lead to te successful prediction of te maturity of oter unknown ceeses. Media simulatin ceese ave been successfully used witin a sort period of time due to rapid deradation. Microoranisms rowin on ceese media were detected after days at o C and verified after days at te same temperature before visual rowt of te microoranisms. Differentiation of yeasts, and filamentous funi on ceese media wit te use of electronic nose tecnoloy was possible after days at o C and for some of te species studied. Identification models could not be used in order to identify te contaminant rowin on te plates. Tis was due to te canes tat occurred to te media makin te lon-term identification impossible. Optimizin te use of sensor arrays wit te elp of multivariate analysis metods was investiated wit te elp of te software packae Unscrambler (CAMO) version 7.6 SR. More stable media appropriate for lon-term identification use of spoilae microoranisms are required. Next stae in tis investiation will be performed wit a new ae ceese media tat is more stable. ACKNOWLEDGMENTS Te researc work performed in tis study is sponsored by te EU project Rapid detection of microbial contaminants in food products usin electronic nose tecnoloy (QLK--76). REFERENCES [] Desmazeaud, M. and Coan, T.M. (996). Role of cultures in ceese ripenin, 7- in Dairy Starter Cultures. VCH Publicers, Inc. New York, USA. [] Molimard, P., Lesscaeve, I., Issancou, S., Brousse, M., Spinnler, H.E. (997). Effect of te association of surface flora on te sensory properties of mouldripened ceese. Le Lait, 77, [] van den Tempel T. and Nielsen P.V. (). Effects of atmosperic conditions, NaCl and ph on rowt and interactions between moulds and yeasts related to blue ceese production. International Journal of Food Microbioloy, 7, 9-99.