J Sci Food Agric 1998, 78, 343È348 Use of Digital Aroma Technology and SPME GC-MS to Compare Volatile Compounds Produced by Bacteria Isolated from Processed Poultry* Judy W Arnold¤ and Samuel D Senter United States Department of Agriculture, Richard B Russell Agricultural Research Center, PO Box 5677, Athens, Georgia 30604, USA (Received 2 March 1998 ; accepted 10 March 1998) Abstract : Digital aroma technology, solid-phase micro-extraction (SPME) and gas chromatographic mass spectral (GC-MS) analysis of the headspace volatile organic compounds were used to compare bacterial species important for food safety and common to bioÐlms in the poultry processing environment. The instrument for digital aroma technology, called the electronic nose, measured changes in resistance of polymer sensors caused by volatile gases from the headspace of samples. Graphical output by the Sammon mapping technique produced patterns of di†erences or similarities among the samples. ArtiÐcial neural network software was used to model groups of samples and classify subsequent unknowns. Compounds isolated from the headspace of sealed cultures using polydimethylsiloxane SPME Ðbres and identiÐed by GC-MS analyses were predominantly alcohols and indole. These qualitative proÐles were repetitive for speciÐc organisms in relation to purity and repeatability of the cultures, di†ered by species and were used as objective standards to compare the graphical outputs of the electronic nose. ( 1998 Society of Chemical Industry. J Sci Food Agric 78, 343È348 (1998) Key words : bioÐlms ; poultry ; bacteria ; aroma ; electronic nose ; SPME ; GC-MS INTRODUCTION volatile organic compounds (VOCs) as they cross an array of sensors (Hodgins and Simmonds 1995). Rapid reversibility of the volatile to sensor binding process allows samples to be run in rapid succession. In early models, sensor arrays usually were connected to numerical modules for statistical analysis, ie semiconductor gas sensors responded to the aromas, pattern recognition analysis di†erentiated the resulting data matrices and cluster analysis was used to classify the samples (Aishima 1991 ; Newman 1991 ; Persaud 1991). A system for odour classiÐcation that was introduced later included a sensor array, statistical analysis module and an artiÐcial neural network. The network classiÐed and stored the statistical data, then used the information to match subsequent unknown samples with previously encountered samples (Dodd et al 1991 ; Shurmer and Gardner 1992 ; Davide et al 1994 ; Hines and Gardner 1994). Digital aroma technology, or the electronic nose, has been developed in recent years as a quality assurance tool to detect odours in food and Ñavour industries. Intended to mimic biological olfactory functions of human sensory panels, it functions by rapidly adsorbing and desorbing volatiles at the surfaces of sensors, causing changes in measured electrical resistance. While analytical techniques, such as gas chromatography (GC), separate headspace gases from samples into peaks corresponding to individual compounds, the electronic nose integrates measurements of the total headspace * This article is a US Government work, and, as such, is in the public domain in the United States of America. Reference to a company name or product does not imply endorsement by the US Department of Agriculture. ¤ To whom correspondence should be addressed. 343 ( 1998 Society of Chemical Industry. J Sci Food Agric 0022È5142/98/$17.50. Printed in Great Britain J W Arnold, S D Senter 344 Historically, qualitative assessment of spoilage potential for foods and the by-products of spoilage has been made by human panelists (Dalgaard 1995) who are presented serial dilutions of air samples for olfactory measurements (Hobbs et al 1995). Quantitative characterisation was then made with model substrates and measurement of speciÐc VOCs by analytical procedures that characterise the aromas or odours. Examples are the isolation and identiÐcation by GC mass spectroscopy (GC-MS) of key VOCs that contribute to the odour of stewed beef (Guth and Grosch 1994) and the analyses that chemically proÐled the major odour compounds of pig and chicken livestock wastes (Hobbs et al 1995). In the latter study, an electronic nose was used to discriminate odours through sensor response patterns. Pattern recognition by the electronic nose also has been used to determine food quality and di†erentiate between raw food materials such as meat, grains and types of cheese (Pisanelli et al 1994 ; Brewer and Vega 1995 ; Jonsson et al 1997 ; Muir et al 1997) and to predict storage time and quality of ground beef (Winquist et al 1993). Analysis of vapours produced by microorganisms involved in sausage fermentation indicated that the method could be used to assess contamination of food products (Rossi et al 1995). The purpose of the present work was to evaluate the speciÐcity, sensitivity and reproducibility of the electronic nose when used to characterise and proÐle headspace VOCs from bacterial cultures isolated from broiler carcasses and grown on deÐned media. The species chosen were representative of potential pathogens commonly found in bioÐlms from the food processing environment. The inÑuence of parameters for bacterial growth, such as medium, temperature and phase of growth were considered and related to changes in the concentration of VOCs as determined objectively by solid-phase micro-extraction (SPME) GC-MS analyses. EXPERIMENTAL Sample preparation The source and names of bacterial species used in this study are shown in Table 1. Field isolates were obtained from whole carcases collected from a commercial broiler processing plant after the New York rinse. Each carcass was rinsed with 100 ml phosphate-bu†ered saline. Aliquots of the rinse were diluted with trypticase soy broth (TSB) in duplicate tenfold series, 1 ml of each plated on plate count agar (PCA) and the plates incubated for 18 h at 37¡C. Isolates were selected for identiÐcation after multiple passages on PCA. For long-term maintenance, aliquots of bacterial cultures were stored at [30¡C in TSB containing 5% glycerol. To resuscitate, an aliquot was thawed and 200 ll was added to 10 ml TSB, then incubated at 35¡C for 18 h, unless otherwise stated. Sensitivity tests of the instruments were performed on tenfold dilution series of 18-h cultures, using TSB, phosphate-bu†ered saline (PBS) or water as the diluent in separate experiments. Sample analysis by digital aroma technology A 5-ml aliquot of the incubated culture was transferred to a 750-ml capacity, single-use, disposable pouch, which was then Ðlled with air that had been Ðltered through indicating Drierite, (WA Hammond Drierite TABLE 1 Bacterial species used in cultures to evaluate head space volatile organic compounds by the electronic nose and gas chromatography mass spectral analysis Species Source % ConÐdencea Cells ml~1 b Salmonella enteritidis Escherichia coli L isteria monocytogenes Klebsiella pneumoniae Enterobacter cloacae Pseudomonas aeruginosa Acinetobacter calcoaceticus ATCCc ATCC ATCC Field isolated Field isolate Field isolate Field isolate 99 99 99 96 99 99 99 2É91 ] 108 2É66 ] 108 1É08 ] 108 3É66 ] 108 4É99 ] 108 4É99 ] 108 3É53 ] 108 a Level of conÐdence for identiÐcation of experimental cultures, performed on a Vitek instrument (bioMerieux, Inc, Hazelwood, MO, USA). b Samples incubated in trypticase soy broth 18 h at 35¡C prior to testing. Each concentration shown was the mean of four experiments. c American Type Culture Collection (ATCC), Rockville, MD, USA. IdentiÐcation of experimental cultures were veriÐed by Vitek. d The method for selecting and maintaining Ðeld isolates is described in the Experimental section. Comparison of volatile compounds produced by bacteria from poultry Co, Xenia, OH, USA). Pouches were stored in a 35¡C conditioning cabinet for 5 min for temperature equilibration and then sampled with the AromaScanner bench-top instrument (Foss Food Technology Corp, Eden Prairie, MN, USA). Temperature and humidity were monitored during sampling, and the relative humidity of sample headspace and reference air were controlled to 16% to allow consistent headspace generation. Between samples, the vapours from a wash bottle containing approximately 20 ml of a 2% butanol solution were sampled for 10 s as a sample line wash. A control of TSB was run as the Ðrst and last sample each day. Data analysis for digital aroma technology The combined responses from the sensors were used by the AromaScanner software to generate a pattern characteristic of the controlled headspace from each sample. The patterns were displayed as line graphs or histograms and stored in databases for later comparison. Multiple discriminant analysis reduced the data for each species into two dimensions viewed in a single plot. The patterns, or proÐle data, were analysed by cluster analysis (Sammon mapping), and then plotted as a map. The axis of each map was in units of Euclidean Distance (ED) (Persaud 1992), which is a measure of di†erences between samples : the greater the ED, the greater the di†erence. An ArtiÐcial Neural Network (ANN) classiÐed and retained the data for comparison with subsequent unknowns. Succeeding samples were classiÐed by reference to the database of previously encountered patterns, and correlation was displayed as a percentage conÐdence level. SPME extraction, GC analysis of headspace volatiles For each species evaluated, two culture tubes (150 ] 16 mm), each containing 10 ml TSB, were inoculated with 200 ll of the speciÐc culture and incubated for 18 h at 35¡C. Two control tubes of TSB were prepared and incubated similarly. After incubation, aliquants from one tube were added to the second to bring the depth of the culture to 110 mm, which resulted in consistent headspace volume among samples. Thorough mixing was succeeded by capping with a 20-mm silicone septa, which was covered with ParaÐlm to assure adequate sealing. The tubes were then placed in a 37¡C water-bath for 30 min before sampling for headspace VOC equilibration. A 100-lm polydimethylsiloxane (PDMS), manually operated SPME Ðbre and holder system (Supelco Inc, Belfonte, PA, USA) was used for headspace sampling. These Ðbres are a recent innovation that have been used for the extraction of VOC from the headspace of such diverse biological materials as Ñavour volatiles from 345 fruit beverages (Penton 1996), forensic and pharmaceutical samples (Mindrup 1995), volatiles from soil and water samples (Schumacher 1997), organic compounds from environmental samples (Zhang and Pawliszyn 1993) and the volatile compounds from human breath (Grote and Pawliszyn 1997). The PDMS Ðbre was inserted through the spectrum of the sample tube and allowed to equilibrate with the headspace volatiles for 30 min. The Ðbre was then retracted into the barrel of the syringe and immediately inserted into the injector of the GC for 1 min desorption of the entrapped VOCs. Chromatographic Ñame ionization detection (GCFID) analyses of the VOCs were performed with a Perkin-Elmer Autosystem XL gas chromatograph (Perkin-Elmer, Norwalk, CT, USA) that was equipped with a 60 m ] 0É25 mm (id), 0É25-lm Ðlm thickness and DB-1 column (J & W ScientiÐc, Folson, CA, USA). Helium was the carrier gas at 30 psig, which resulted in a linear velocity of 30 cm s~1 with the column at 40¡C. Chromatograph operating conditions were as follows : initial oven temperature 40¡C with no hold time, then programmed to 275¡C at 8¡C min~1 and held for 5 min ; injector temperature 250¡C ; detector temperature 310¡C. The injector was held in the splitless mode for 1 min after insertion of the Ðbre ; the purge was then automatically engaged. GC-MS analyses Qualitative analysis of the extracted volatiles was performed with a Finnigan MAT GCQ mass spectrometer (Finnegan, San Jose, CA, USA) operated in the electron-ionisation (EI) mode. Electron energy was 70 eV, multiplier voltage 1100 V, source temperature 200¡C and transfer line 240¡C. Spectral data was acquired over a mass range of 28È200 amu at a scan rate of 0É6 s scan~1. Chromatographic conditions for the interfaced GC were as follows : column 60 m ] 0É25 cm (id), 0É25-lm Ðlm thickness DB-1 (J & W ScientiÐc) using He carrier gas at 28 cm s~1 linear velocity at 40¡C ; oven temperature initially 40¡C with 1 min hold, then programmed to 275¡C at 8¡C min~1 and held for 5 min ; injector temperature 225¡C. The SPME absorption and desorption time was the same as in the preceding chromatographic conditions, with the injector (0É75-mm liner) operated in the splitless mode for 1 min, then in the purge mode for the remainder of the run. RESULTS Detection Analysis of the VOCs in the headspace of samples tested by digital aroma technology detected the presence of each of the bacterial species tested : Salmonella J W Arnold, S D Senter 346 enteritidis, Escherichia coli, L isteria monocytogenes, Klebsiella pneumoniae, Enterobacter cloacae, Pseudomonas aeruginosa and Acinetobacter calcoaceticus (Table 1). After data was collected for multiple samples of each species tested, the electronic nose was able to distinguish each of the cultured species from each other in subsequent analysis. Responses to the di†erent samples and mapped data are shown in Fig 1. Di†erences or similarities between the bacterial species are shown by the relative spatial separation of the data point clusters. Fig 1. Mapping of the response patterns generated by the electronic nose for bacterial species. Samples were incubated in TSB, 18 h, 35¡C prior to testing. Each point represents the data taken from separate experiments performed on nine different days : (=) Salmonella enteritidis ; (…) Escherichia coli ; (>) Enterobacter cloacae ; (@) Pseudomonas aeruginosa. For each sample, the Sammon mapping technique reduced the responses from the 32 sensors to a single data point and compared the points by non-linear dimensional analysis. Stability Sample response patterns were reproducible over time. A database, generated for each species, contained patterns from samples tested on each of nine di†erent days (Fig 1). The responses to the sensors of each species were stable over the course of the experiment. Multiple classes or descriptor levels of bacterial odour proÐles were processed and retained by the ANN for comparison of new samples with previously encountered samples. Sensitivity The sensitivity and discriminating ability of the ANN increased as more data points were collected for the standard samples. Sensitivity of the system was further tested by a dilution experiment in which the bacterial cultures were diluted in a tenfold series of TSB, PBS and water. When data from the samples shown in Fig 1 were used to train the ANN and diluted samples were compared, recognition was achieved. Chromatographic analyses Fig 2. Comparative gas chromatograms of volatile organic compounds (VOCs) from the headspace of Escherichia coli and Pseudomonas aeruginosa cultures using polydimethylsiloxane (PDMS), solid-phase micro-extraction (SPME) Ðbres (Supelco Inc, Bellfonte, PA, USA) for extraction and concentration. B refers to the VOCs isolated from the culture broth, and the numerals refer to the identiÐed bacteria VOCs listed in Table 2. Gas chromatograms of the VOCs isolated from the headspace of E coli and P aeruginosa are presented in Fig 2 and represent the diversity in the headspace VOCs isolated from cultures of the eight species exposed to the PDMS Ðbres. Compounds identiÐed from the isolates are presented in Table 2 as area percentage of the chromatograms with threshold level of the PE-Nelson Turbochrom-4 integration system (PerkinÈElmer) set at 500 lV s~1. IdentiÐed compounds were predominantly alcohols, with ethanol being the major constituent in all species except P aeruginosa and A calcoaceticus. In these species, components characteristic of the other species were absent except for P aeruginosa, which produced only 3-methyl-1-butanol and phenethyl alcohol in measurable quantities. Species producing the greater number of compounds, predominantly alcohols, were S enteritidis, E coli, K pneumoniae and E cloacae. L isteria monocytogenes was distinguished Comparison of volatile compounds produced by bacteria from poultry 347 TABLE 2 Volatile organic compound isolatesa from the headspace of bacterial cultures from processed poultry by polydimethylsiloxane, solid-phase micro-extraction (comparative areas of the chromatograms) Peak 1 2 3 4 5 6 7 8 9 10 11 12 RT b 3 : 76 4 : 20 4 : 98 6 : 07 12 : 92 13 : 68 16 : 89 17 : 10 17 : 22 20 : 84 23 : 87 24 : 17 Compound Ethanol 1-Propanol 3-Methyl butanal 3-Methyl-1-butanol Octanol Phenethyl alcohol 9-Decene-1-ol Decanol Indole Dodecanol cis-7-Tetradecene-1-ol Tetradecanol Bacteria speciesc and percent aread SE EsC LM KP EnC 47É9 2É6 25É7 2É7 12É6 63É9 1É8 40É8 0É8 3É0 1É6 0É5 2É6 4É8 1É3 0É6 3É6 0É7 10É6 41É2 1É9 0É6 0É5 0É8 6É3 12É9 1É2 1É3 0É7 6É1 3É9 0É7 2É4 5É2 2É2 2É6 3É2 2É2 2É7 4É6 0É3 0É4 1É7 PA AC 1É6 3É2 a IdentiÐcation by comparison of mass spectra and chromatographic retention times with authenic standards. b Retention time of the individual compounds identiÐed. c SE, Salmonella enteritidis ; EsC, Escherichia coli ; LM, L isteria monocytogenes ; KP, Klebsiella pneumoniae ; EnC, Enterobacter cloacae ; PA, Pseudomonas aeruginosa ; AC, Acinetobacter calcoaceticus bio anitratus. d Percentage area of the chromatogram. by the production of 3-methyl butanal and E coli by the production of indole. Extractable volatiles from the headspace of TSB were predominantly pyrazine derivatives and benzaldehyde. DISCUSSION Because traditional human evaluation of odour is very subjective, digital aroma technology should lend itself well to the development of a system based on objective measurements. The use of human odour panels to evaluate and control the quality of raw materials for spoilage, or bacterial contamination of Ðnished products, is labour-intensive, time-consuming, expensive, prone to errors and can be hazardous to human health. Objective odour measurement for improving quality control from raw material to Ðnished product could provide real-time detection of process problems. Early detection and measurement of hazardous or contaminated samples would have cost-saving beneÐts for raw material quality control, new product development and Ðnal product standardisation. Digital aroma technology can alleviate these concerns. Results showed that digital aroma technology was sufficiently sensitive to measure the volatiles in the headspace of bacterial samples, and the method was reproducible within the limits of the analyses. Changes in sensor resistance generated data to cluster and group samples. The array of sensors had di†ering degrees of selectivity towards di†erent volatile compounds, as evidenced by the GC-MS analyses where substantial diversity was observed by species. Samples were processed rapidly, and data was output in graphical form and stored to later compare and classify succeeding samples. The sensitivity and discriminating ability of the instrument was improved as more data points were collected and further processed. Neural network algorithms performed real-time recognition ; conÐdence levels were reported for subsequent data validation. It was not the intention of this work to discover new VOCs produced by known species of bacteria in a food processing environment, but to detect and proÐle headspace VOCs by a new analytical technique (the electronic nose) and conÐrm observed di†erences by a new isolation technique (SPME) and standard analytical procedures (GC-MS). Selectivity of the PDMS Ðbres used was a limiting factor in the extraction of these volatiles for GC-MS analyses ; however, the availability of di†erent Ðbres has been improved since the completion of this study and may o†er an improved approach to this type of analysis. These experiments suggest that digital aroma technology may have potential for studying the VOCs of bacteria. However, the aroma technology system needs further development to improve reproducibility and sensitivity and to reduce the time required for the experimental work and data handling. The preceding results acknowledge di†erences in the headspace VOCs that constitute the complex odours J W Arnold, S D Senter 348 generated by bacteria and shows that a sensory study on the basis of digital aroma technology, static headspace analysis with SPME GC-MS analysis is a promising approach to detect the key compounds and bacterial species implicated in food spoilage and is important for food safety. ACKNOWLEDGEMENTS The authors wish to thank T Breedlove, P Mason and K Tate for technical assistance. REFERENCES Aishima T 1991 Discrimination of liquor aromas by pattern recognition analysis of responses from a gas sensor array. Anal Chim Acta 243 293È300. Brewer M S, Vega J D 1995 Detectable odor thresholds of selected lipid oxidation compounds in a meat model system. J Food Sci 60 592È595. Dalgaard P 1995 Qualitative and quantitative characterization of spoilage bacteria from packed Ðsh. Int J Food Microbiol 26 319È333. Davide FA M, Di Natale C, DÏAmico A 1994 Self-organizing multi-sensor systems for odour classiÐcation : internal categorization, adaptation and drift rejection. Sens Actuators B 18 244È258. Dodd G, Bartlett P, Gardner J 1991 Complex sensor systems : odour detection by the sense of smell and by electronic noses. Biochem Soc T rans 19 36È39. Grote C, Pawliszyn J 1997 Solid-phase microextraction for the analysis of human breath. Anal Chem 69 587È596. Guth H, Grosch W 1994 IdentiÐcation of the character impact odorants of stewed beef juice by instrumental analyses and sensory studies. J Agric Food Chem 42 2862È 2866. Hines E L, Gardner J W 1994 ArtiÐcial neural emulator for an odour sensor array. Sens Actuators B 19 661È664. Hobbs P J, Misselbrook T H, Pain B F 1995 Assessment of odours from livestock wastes by a photoionization detector, an electronic nose, olfactometry and gas chromatographyÈ mass spectrometry. J Agric Eng Res 60 137È144. Hodgins D, Simmonds D 1995 The electronic nose and its application to the manufacture of food products. J Autom Chem 17 179È185. Jonsson A, Winquist F, Schnurer J, Sundgren H, Lundstrom I 1997 Electronic nose for microbial quality classiÐcation of grains. Int J Food Microbiol 35 187È193. Mindrup R F 1995 Solid phase microextraction simpliÐes preparation of forensic, pharmaceutical, and food and beverage samples. T he Reporter (Supelco) 14 2È5. Muir D D, Hunter E A, Banks J M 1997 Aroma of cheese. 2. Contribution of aroma to the Ñavour of Cheddar cheese. Milchwissensch Milk Sci Int 52 85È88. Newman A R 1991 Electronic noses. Anal Chem 63 585AÈ 588A. Penton Z 1996 Flavor volatiles in a fruit beverage with automated SPME. Food T est Anal 2 16È18. Persaud K C 1991 Odour detection using sensor arrays. Anal Proc 28 339È341. Persaud K C 1992 Electronic gas and odour detectors that mimic chemoreception in animals. T rends Anal Chem 11 61È67. Pisanelli A M, Qutob A A, Travers P, Szyszko S, Persaud K C 1994 Applications of multiarray polymer sensors to food industries. L ife Chem Rep II 303È308. Rossi V, Talon R, Berdague J L 1995 Rapid discrimination of Micrococcaceae species using semiconductor gas sensors. J Microbiol Meth 24 183È190. Schumacher T L 1997 Fast screening of water and soil samples using solid phase microextraction (SPME). T he Reporter (Supelco) 16 8. Shurmer H V, Gardner J W 1992 Odour discrimination with an electric nose. Sens Actuators B 8 1È11. Winquist F, Hornsten E G, Sundgren H, Lundstrom I 1993 Performance of an electronic nose for quality estimation of ground meat. Meas Sci T echnol 4 1493È1500. Zhang Z, Pawliszyn J 1993 A by a rapid, new analytical procedure (SPME analysis of organic compounds in environmental samples by headspace solid phase microextraction). J High Res Chrom 16 689È692.