Final Report Summary - SYMBIOSIS-EU (Scientific sYnergisM of nano-Bio-Info-cOgni Science for an Integrated system to monitor meat quality and Safety during production, storage, and distribution in EU)
Executive Summary:
The SYMBIOSIS-EU is a European project that brings together 14 beneficiaries from 6 EU countries (plus one each from NZ and US) to study tools that will allow a rapid evaluation of meat quality and safety in comparison with the time consuming microbiological analyses or the unattractive and expensive sensory analysis, of which both are most often used to evaluate the freshness, spoilage and/or safety of meat and meat products (EU 2005).
The overall aim is to identify and quantitatively evaluate practical and easy to use chemical, biochemical and molecular indices and establish their applicability as quality monitors for inspection of meat safety and quality. The project has applied a multidisciplinary system wide approach relying on converging technologies (bioinformatics, nanotechnology, modeling) to obtain knowledge for meat quality that has been translated into simple devices and practical indicators of quality and safety.
The main objectives met in this project were (i) to develop and/or validate easy to use chemical/biochemical methods (e.g. biosensors, fluorescence, FT-IR), molecular methods (DNA, microarrays), (ii) to develop a suitable software platform for data sharing and integration, (iii) to apply multivariate statistical methods and machine learning (neural networks, fuzzy logic, genetic algorithms) to identify robust multiple compound quality indices, (iv) integration of the sensors and information platform and development of a system to automatically transform data acquired from a sample into a diagnosis of meat safety and quality.
Project Context and Objectives:
Project Description
The current practice to assure the safety/quality of meat still relies heavily on regulatory inspection and sampling regimes. This approach, however, seems inadequate because it cannot sufficiently guarantee consumer protection since 100% inspection and sampling is technically, financially and logistically impossible. In addition, the following issues need to be considered: (i) more than 50 chemical, physical and microbiological methods have been proposed for the detection and measurement of bacterial safety or spoilage in meat, (ii) most of these methods are time-consuming and provide retrospective information and thus they cannot be used on- or at-line, (iii) the lack of general agreement on the early quality changes, and (iv) the changes and development in technologies for food processing and preservation [e.g. vacuum packaging (VP), modified atmosphere packaging (MAP), active packaging, etc] make it evident that the important and urgent task of identifying safety and spoilage indicators is a complicated proposition.
Beyond the Baseline data
The basic concept in this project is related with the integration of the molecular tools (WP1, WP6) as well as sensors /biosensors (WP3) and other technologies (e.g. FT-IR ATR, RAMAN) with the information platform (WP4) and to develop an expert system (M4.1) able to automatically classify the sensorial input into a diagnosis based on extracted pre-processing features (WP4 and WP5). Additionally this project will generate data covering genomic, proteomic, metabolomic, and phenotypic aspects. This, therefore, requires the integration (WP4 and WP5) of highly divergent data generated from the various high throughput analytical techniques such as HPLC, GC-MS, Microarray, Electronic nose, Fluorescence Detectors and multi-spectral imaging (M4.1 M4.3). Due to the heterogeneous nature of the data generated, and the geographic distribution of the project beneficiaries, an important task will be to create a central data (M4.2) repository that will serve as a common framework for the research project.
Specific Objectives per WorkPackage
Main objectives of WP0
The objectives for the reporting period regarding WP0 were: (i) The Overall Project Management (Task 0.1) (ii) To fulfil the general coordinators responsibilities (Task 0.2) (iii) Collate the beneficiaries financial and technical contributions and submit reports to the EU Commission. In the meetings the GA, and GB were also organized (Task 03; Subtask 0.3.1 and Subtask 0.3.2 was achieved last June in Brussels). A revised plan was delivered after the review and the assessment report for the reviewers.
Main objectives of WP1
This WP was finalized in month 30; the main objectives related to the reporting period were: (i) To finalized the identify and characterize ESO associated with the spoilage of meats produced and stored under different temperatures and packaging systems (MAP/VP). (ii) To collect information on the spoilage process of meat that will lead to reliable and specific kinetic spoilage models; both these Tasks 1.1 and task1.2 completed well.
The characterization of microbial association of various meat and meat products has been completed and most of the data have been published or have been submitted for publication.
Main objectives of WP2
Similarly this WP was ended in month 30; during the last 18M period the main objective was: To determine the effect of intrinsic parameters (physical, chemical and structural properties inherent in meat itself) and extrinsic parameters, i.e. temperature imposed atmosphere, microbial association, quorum sensing, on the formation of microbial metabolites by ESO. This was achieved and the corresponding deliverable was provided on time. Most of the results have been published or have been submitted for publication. Additionally this database was subject to WP5, WP7 and has been used successfully to fulfill the achievements of the above mentioned WPs.
Main objectives of WP3
The WP ended in month 30; in the running period the main objective of wp3 was to record all the derived data which should have been used to develop models / tools that have been proved to be able to determine the potential for spoilage of meat products. An additional output will be the information about the potential for such approaches to detect the metabolites of food-borne pathogenic bacteria. The huge data set derived from this WP, includes spectra from vibrational spectroscopy instruments, e.g. FTIR, Raman, surface chemistry as measured with VIDEOMETER, as well as the metabolomics, e.g. data from HPLC, GC/MS and e-nose, were the basis to feed the platform developed by other beneficiary for WP4, WP5 and WP7.
Main objectives of WP4
In the running period the main objectives of wp4 were: (i) To develop a database by capturing all of the experimental data from the project. The database will be specifically designed to cope with the heterogeneous nature of the sensorial data acquired during the project and (ii) To develop an integrated end-user interface to visualize all acquired information/results.
Main objectives of WP5
The main objectives of wp5 were: (i) The evaluation of the sensorial information through on reliability of information, redundancy, and discriminative character; and (ii) The development of a diagnostic system to classify automatically the sensorial input into a decision based on the extracted features.
At the end of the project two more objectives should be completed: (i) The development of modeling tools for meat spoilage, and (ii) Kinetic and prediction modeling.
Main objectives of WP6
The main objective of this wp for the first 18M was: To investigate the expression of key genes involved in synthesis of specific spoilage metabolites under different environmental conditions induced by storage and packaging. The 2nd objective is to investigate the expression of virulence and stress response related genes under different environmental conditions induced by storage and packaging.
Main objectives of WP7
The aims of WP 7 were to evaluate indices of quality and mathematical models for metabolite formation at a European level. In particular, an evaluation of differences between data from laboratories in separate countries will be possible because common analytical methods will be used within the present project (Task 7.1 and Task 7.2). It should be noted that several beneficiaries of the project possess equivalent advanced equipment like GC-MS and HPLC instruments or transportable miniaturized sensors (Task 7.2) that may easily allow on-line data recording in plants in several EU countries. Data from Task 7.1 and 7.2 will be used for validation of mathematical models developed within the project in WP4 and WP5. The produced models will be evaluated at European level (Task 7.3).
Testing the developed techniques, tools and methodologies, validation of the mathematical models in real time (D7.1 D7.3) (M30, M36), will give the opportunity to measure the efficiency of the produced models/tools. The verification of this models/tools (D7.2; M30) can be used to elaborated further the meat spoilage.
Main objectives of WP8
In this WP the last period the main objectives are: (i) To include successfully validated models in application software and thereby facilitate their practical usefulness, and (ii) To establish permanent Training and Dissemination Units at various Meat and other Industries in selected EU countries. Since the development of SORF was proved to be a general tool, this exhibition was done to different industries also.
Main objectives of WP9
The main objective of this wp for the project period was: Communication and transfer knowledge from all WP (Task 9.1). Organisation of training sessions and implementation of distance learning packages (e.g. bioinformatics, molecular tools, chemometrics, mathematical modeling, etc). Training activities (Task 9.1) will deliver both traditional (e.g. residential courses; within Universities, company training days, plant visits; and web-based (Task 9.2) as well as in workshops and conferences.
Project Results:
Effect of Microbial Association on meat Spoilage
Meat is recognized as one of the most perishable foods. Apart from the physical damage, oxidation and color change, the other spoilage symptoms are due to the undesirable growth of microorganisms to unacceptable levels. The microbial quality of meat depends on the physiological status of the animal at slaughter, the spread of contamination during slaughter and processing, the temperature and other conditions of storage during distribution (Nychas et al. 2008). When large numbers of undesirable microorganisms are present in raw meat, then there will be undesirable changes such that the food becomes unappealing and unsuitable for human consumption (Nychas et al. 2008). Many groups of organisms contain members potentially contributing to meat spoilage under appropriate conditions. This makes the microbial ecology of spoiling raw meat very complex and thus the spoilage very difficult to prevent.
Development of microbial association
Within the SYMBIOSIS EU project attention has been also given to the characterization of the spoilage microbiota not only at species but also at strain level. This is in fact believed to potentially play a pivotal role in the establishment of the meat spoilage as different strains of the same species may have different spoilage activities or can be differently affected by antimicrobial storage conditions (Andritsos et al. 2012; Casaburi et al., 2011; Doulgeraki et al. 2011; Ercolini et al., 2010b). Therefore, characterization of lactic acid bacteria (Doulgeraki et al. 2010), Enterobacteriaceae, Br. thermosphacta, Pseudomonas spp., and Carnobacterium spp. (Casaburi et al., 2011; Doulgeraki et al. 2011; Ercolini et al. 2010b; Doulgeraki and Nychas 2012 ; Papadopoulou et al. 2012) isolated during storage of meat under different conditions (aerobic, vp, MAP with or without essential oils) have been recently performed. Conventional and molecular tools that have been used in this project for detection and identification of the bacteria are also presented and discussed in a recent publication (Doulgeraki et al. 2012).
Succession of specific groups
With regard to Lactic Acid Bacteria (LAB), a total of 266 strains have been isolated from minced beef stored at 0, 5, 10 and 15o C aerobically and under modified atmosphere packaging consisting of 40% CO2-30% O2-30% N2 in the presence MAP (+) and absence MAP (-) of oregano essential oil. Sequencing of their 16S rRNA gene along with presence of the katA gene demonstrated dominance of the LAB microbiota by Leuconostoc spp. during aerobic storage at 5, 10 and 15o C as well as during MAP (-) and MAP (+) storage at 10 and 15o C Lactobacillus sakei prevailed during aerobic storage at 0 as well as at MAP (-) and MAP (+) storage at 0 and 5o C. The sporadic presence of other species such as Leuconostoc mesenteroides, Weisella viridescens, Lactobacillus casei and Lactobacillus curvatus has also been determined. Pulsed-Field Gel Electrophoresis of high molecular weight genomic DNA revealed the dynamics of the isolated LAB strains. Prevalence of Leuconostoc spp. was attributed to one strain only. On the other hand, packaging conditions affected Lb. sakei strain spoilage dynamics.
Effect of natural inhibitors on pathogenic bacteria
Several studies (Argyri et al. 2010; Ferrocino et al 2012; La Storiaet et al. 2008) in this project have demonstrated the inhibitory effect of volatiles from Oregano Essential Oil against the microbial association of meat either beef or pork. The results have been mentioned above. Additional studies have reported the effect of 3 bacteriocins and this was exploited with the atomic force microscopy (AFM) analysis for the characterization of plastic films activated with antimicrobial agents. The aim was to acquire information on the distribution of the antimicrobials on plastic materials with the ultimate scope of understanding the mechanisms of interaction between antimicrobials and materials to be used for food packaging. Four polyethylene films differing in linear, EVA, and erucamide content were activated by 3 different bacteriocins as antimicrobials, namely, nisin and bacteriocins Bac162W from Lactobacillus curvatus and BacAM09 from Lactobacillus plantarum. The spectrum of activity of the bacteriocins was assayed and shown to include several strains of Listeria monocytogenes. The plastic films were activated by a previously developed coating procedure and the surfaces of the active films were examined by AFM. In addition, roughness parameters related to the single surfaces were investigated by appropriate software.
Contribution of Energy Sources and microbial metabolites on meat spoilage
It is well established that low molecular weight compounds, e.g. glucose, lactic acid and certain amino acids are catabolized by almost all the bacteria of the meat microflora, followed by nucleotides, urea and water-soluble proteins. The former compounds are the essential energy sources for the massive growth of the microcosm on the meat despite their negligible quantity in comparison to proteins. It is shown that actual concentrations of these compounds can affect the type (e.g. saccharolytic, proteolytic) and the rate of spoilage, and moreover they seem to be the principal precursor(s) of those microbial metabolite(s) that we perceive as spoilage (Nychas et al. 2008).
There are three classes of substances that are used by the microbial association: (i) compounds involved in the glycolytic pathway (e.g. glycogen, glucose, lactate, etc.), (ii) metabolic products (e.g. gluconate, pyruvate, lactate, etc), and (iii) nitrogen energy sources (e.g. aminoacids, proteins). Glucose has been found to be the precursor of many off-odours during meat storage (Nychas et al. 2008), while its limitation could cause a switch from a saccharolytic to an amino acid degrading metabolism in at least some bacterial species. The changes of glucose and lactate as well as their oxidative products (e.g. gluconate) have been proposed for use in describing or predicting the degree of spoilage (Nychas et al. 2008). The information, about the rate and the sequence of utilization of available substrates in space and time, by the ESO in meat stored under different (dynamic) conditions, is rather negligible. The nutrient contribution may be related to positive (synergistic/syntrophic) or competitive activities for nutrients/energy (e.g. under excess, limitation or starvation), while metabiosis (production of a favorable environment) and cell-to-cell communication (quorum sensing) could also affect the physiological attributes of the organisms under the imposed ecological determinants.
In a specific study, beef chops were stored at 4°C in different conditions: air (A); modified atmosphere packaging (MAP); vacuum (V); bacteriocin-activated antimicrobial packaging (AV). After 0 to 45 days of storage, analyses were performed to determine loads of spoilage microorganisms, microbial metabolites by SPME-GC/MS and 1HNMR, microbial diversity by PCR-DGGE and pyrosequencing.
The metabolic profile of minced beef stored (i) aerobically, (ii) under modified atmosphere packaging (MAP), and (iii) under MAP with oregano essential oil (MAP/OEO) at 0, 5, 10, and 15 °C was investigated suing HPLC and GC/MS (Argyri et al. 2011,2011c). The analysis of chromatograms from HPLC resulted in the selection of 17 discrete peaks (with purity greater than 99%). Fifteen (15) of them are presented in all samples whilst the peaks with RTs of 20.5 and 28.1 min were evident only when spoilage was very pronounced. From the metabolomics point of view, the information contained in all peaks is critical for the characterization of spoilage. In this respect, the whole profile (all the peaks) as derived from spectral data was taken into consideration. Since some of the peaks were not identified and thus concentration values could not be calculated for these peaks, the analysis was continued using the chromatographic area of the peaks. The results demonstrated changes in the chromatographic areas under the peaks of the eluted acids, that were associated or not with storage conditions (e.g. temperature, packaging).
Development and application of vibrational spectroscopy and image /sensor based platform to monitor meat quality
Since the evaluation of the degree of meat spoilage and/or hygiene level is usually made either subjectively based on qualitative criteria or by time consuming, destructive microbiological analyses that do not provide the â??immediate answer requiredâ?? (Nychas et al. 2008). Some approaches are based on the use of Vibrational Spectroscopy, which has the advantage over infrared since it is non-invasive, non-destructive, and the sampling protocols are simplified. This is because it can simply analysed through glass containers, and the sample may be simply analysed in its raw form without dilution. Instruments of vibrational spectroscopy methods among others are FT-IR, Raman spectroscopy.
On the other hand, electronic noses are instruments used to classify and identify complex odours since they reproduce structure and principles of the natural olfactory sense. Basically the electronic nose is composed of an array of sensors and a data analysis system. Sensors act like biological receptors and data analysis system allows to transpose information that sensors extract from an odor in an olfactory image analogous with our sensation of a smell. The identification and comparison of odors is possible because different odors have different olfactory images. This distinguishes the Electronic Nose from other analytical techniques, e.g. gas chromatography which identifies and measure single molecular compounds inside a gaseous mixture (Papadopoulou et al. 2011a, 2012).
The main findings related to the mentioned techniques are summarised below.
The results showed that electronic nose, FTIR, Raman Spectroscopy, surface chemistry as expressed by the multispectral instrument and fluorescent spectroscopy could offer a powerful and rapid tool (i) for quality evaluation of fresh meat (beef, pork) e.g. organoleptic shelf life, (ii) for the measurement of the population of different microbial groups, and (iii) can be applied for the prediction of the microbial loads of the different microbial groups.
The FTIR spectral data collected, were analysed and correlated with the microbiological data and the sensory scores in order to build qualitative and quantitative models that predict the spoilage status of an unknown meat sample. A number of studies, being the outcome of this Project, have demonstrated the potential use of this non-invasive technique (Argyri et al. 2009, 2010; Brewster and Goodacre 2012, Nicolaou et al. 2011). In these articles, results have been analysed with Principal Components Analysis (PCA) and Factorial Discriminant Analysis (FDA), partial least squares regression (PLS-R) models and Support Vector Machine (SVMs) models. Moreover, different pre-processing methods applied have improved the performance of the models (Xu and Goodacre 2011, Argyri et al. 2012).
The RAMAN spectra were introduced to minced beef which was stored aerobically and under modified atmosphere packaging (40%CO2/ 30%O2/ 30%N2) at 5°C and microbiological analysis were performed in parallel with pH measurements, and sensory analysis. Moreover, minced beef samples stored aerobically and under the same modified atmosphere with and without an oregano essential oil (OEO) (2% v/w) slow release system at four temperatures (0, 5, 10 and 15°C), where analysed with Raman spectroscopy. The data derived from the microbiological analysis (Total Viable Counts, Pseudomonas sp., Brochothrixthermosphacta, lactic acid bacteria (LAB), Salmonella Enteritidis) were highly correlated with the Raman spectra (Argyri et al. 2012).
The portable fluorescence spectrometer device equipped with an excitation LED, a fibre optic probe and a mini-spectrometer was tested to quantify minced beef meat spoilage during storage and provided an excellent prediction accuracy of the regression models obtained for meat stored under vacuum and aerobic atmosphere.
Similar results for prediction of microbial flora on minced beef were obtained from portable spectrofluorometer (Kaddou et al. 2011).
The feasibility of Fourier transform infrared (FT-IR) spectroscopy to quantify biochemical changes occurring in fresh minced pork and beef meat in the attempt to monitor spoilage has been applied in this project. A number of articles have been already published (Papadopoulou et al. 2011b; Panagou et al. 2011). For this reason, partial least squares (PLS) models and ANN were constructed to correlate spectral data from FTIR with minced pork meat spoilage during aerobic storage of meat samples at different storage temperatures (0 to 20°C). Spectral data were collected from the surface of meat (both pork and beef) samples in parallel with microbiological analysis to enumerate the population of total viable counts, Pseudomonas spp., Brochothrix thermosphacta, lactic acid bacteria and Enterobacteriaceae. Qualitative interpretation of spectral data was based on sensory evaluation, using a three point hedonic scale, discriminating meat samples in three quality classes, namely fresh, semi-fresh and spoiled. The purpose of the developed models was to classify minced pork samples in the respective quality class, and also to correlate the population dynamics of the microbial association with FTIR spectra. The obtained results demonstrated good performance in classifying meat samples in one of the three predefined sensory classes. The overall correct classification rate for the three sensory classes was 94.0% and 88.1% during model calibration and validation, respectively. Furthermore, PLS regression models were also employed to provide quantitative estimations of microbial counts during meat storage.
B-E- nose
The performance of a quartz microbalance based electronic nose has been evaluated in monitoring aerobically and under MAP packaged beef fillets spoilage at different storage temperatures (0 to 20°C). Electronic nose data were collected from the headspace of meat samples together with microbiological analysis for the enumeration of the population dynamics of total viable counts, Pseudomonas spp., Brochothrix thermosphacta, lactic acid bacteria and Enterobacteriaceae. Qualitative interpretation of electronic nose data was based on sensory evaluation discriminating samples in three quality classes, i.e. fresh, semi-fresh, and spoiled. Support Vector Machines (SVM) classification and regression models using radial basis kernel function were developed to classify beef fillet samples in the respective quality class, and also to correlate the population dynamics of the microbial association with sensors responses. The obtained results demonstrated good performance in discriminating meat samples in one of the three pre-defined quality classes.
C-Fluorescence Spectrometer
Kaddour et al. (2011) used a portable fluorescence spectrometer to quantify minced beef spoilage. This study was investigated on samples stored aerobically and under vacuum at 5 and 15 °C. Total viable counts (TVC), Pseudomonas, lactic acid bacteria, and yeast/molds counts were investigated with classical culture methods. Fluorescence spectra were recorded on the same samples using different excitation LEDs (280, 320, and 380 nm). PLS-R with leave-one-out cross validation was used to perform calibration and validation models. The PLS-R models presented good R2CV (0.50-0.99) and ratio of standard deviation (RPDCV: 1.40-8.95) values after cross-validation. It could be concluded that portable spectrofluorimeters are promising devices to evaluate spoilage in minced beef.
D-Image analysis
The rapid multispectral imaging device was used to monitor quality and to quantify the degree of spoilage of stored minced pork meat (Dissing et al. 2012). Bacterial counts of a total of 155 meat samples stored for up to 580 hours have been measured using conventional laboratory methods. Meat samples were maintained under two different storage conditions - aerobic and modified atmosphere packages as well as under different temperatures. Besides bacterial counts, a sensory panel has judged the spoilage degree of all meat samples into one of three classes. Results showed that the multispectral imaging device was able to classify 76.13% of the meat samples correctly according to the defined sensory scale. Furthermore, the multispectral camera device was able to predict total viable counts with a standard error of prediction of 7.47%. It is concluded that there is a good possibility that a setup like the one investigated will be successful for the detection of spoilage degree in minced pork meat.
E-Raman Spectroscopy
The results are analysed in the validation section.
Understanding Regulation of Metabolomics for Meat Quality
A number of studies were performed to investigate the expression of key genes involved either in synthesis of specific spoilage metabolites and on virulence and stress response under different environmental conditions induced by storage and packaging. In brief, transcriptomic analyses related to the regulation and control of microbial metabolite profiles of Pseudomonas during meat spoilage under various storage and packaging conditions was performed. In addition, transcriptomic analysis related to genes involved in virulence and stress response in Salmonella during exposure of meat to various storage and packaging was performed.
In the case of Salmonella, the stress response of Salmonella Thompson the exposure of sublethal concentrations of thymol (0.01%) at 37°C was studied (Di Pascua et al. 2010). Emphasis was given in the expression of Thymol, which is a natural biocide and component of some essential oils from herbs. Although it's inhibitory effect on the growth of different microorganisms is well documented, the precise targets of the antibacterial action of thymol is not yet been fully established, the action seems to take place in different ways. The strain Salmonella enterica serovar Thompson MCV1 was grown in the presence of a sublethal concentration (0.01%) of thymol. The proteins extracted from treated and untreated cells were subjected to 2-D PAGE, followed by in-gel spot digestion and subsequent MALDI-TOF analysis. In parallel, control and exposed S. Thompson cells were collected and the gene expression of several virulence genes by qPCR analysis, i.e. hilA, stn and sopB was studied. The analysis of gels showed many proteins that were either upregulated or downregulated by the presence of thymol, with significant changes in proteins belonging to different functional classes. In particular, the thioredoxin-1 was not expressed in the treated cells, indicating that its absence could be a consequence of the stress caused by the presence of thymol. On the other hand, different chaperon proteins were upregulated or de novo synthesis such as GroEL and DnaK, key proteins in the protection mechanism toward thermal stress. Outer membrane proteins were upregulated in treated cells; indeed the bacterial envelope stress response is trigged by the accumulation of misfolded outer membrane proteins.
Experimental dataset file upload system
Before uploading an experiment dataset, a new experiment instance has to be defined within the system. This is done through a Web interface page for defining an experiment name, description, type and data (See Figure 1a). Once these fields are filled, the user can then navigate to the next page where he is provided with a file upload system. The experiment type specified by the user in the first navigation page specified the file upload fields present in this step.
Integration of multi-sensor devices output
The design of the SORF back-end has been extended to integrate the experimental output of many sensorial platforms and experimental data into one experiment entity. One experiment stored with the Symbiosis system can have several dimensions of datasets. These datasets can be coming from heterogeneous experimental platform.
System overall architecture
The design and architecture of the Microstudio platform (Now called The Symbiosis Online Research Framework â?? SORF) has gone through various stages of development, as new technolpogies have been made available throughout the timeline of the Symbiosis-EU project (Mohareb and Bessant 2009, Mohareb 2011a,b). The initial front-end design was implemented using Java Enterprise 6 and JSP, deployed on Sun Application Server 8. This was then replaced at later stages of development by the Java Server Faces 2.0 technology, a more advanced and robust framework that allowed us to provide a more intuitive and interactive interface to the system users.
The Data Layer
The Symbiosis MySQL database is a crucial part of the SORF web application. Its main purpose is the user, experiments, datasets and data management. The Symbiosis Database is composed of two parts. The first part is responsible for storing user information such as name, contact details and organization as well as user's publications. The second part is responsible for storing experiments and their datasets. This work focuses only on the usage of the second part of the Symbiosis Database. The idea behind the Symbiosis Database is that the user may create experiments. The experiments may contain datasets whereas each dataset contains data of certain type.
The browse experiments page
Experiments are available in a form of a tabular view. The experiment table allows interactive searching for a particular entity using ajax-based engines by searching for experiments using experiment name, description, WP name, etc. The Steps involved in adding a new experiment to the system remain unchanged as described in the Deliverable Report 4.1 however, a slight modification of the interface has been made.
The analysis framework pages:
The Experiment and Dataset Details tool is composed of two parts: Experiment details and Dataset details which functionality will be briefly described in the sections as follow.
Depending on the metadata dataset in the selected experiment, the tool produces a pie chart for the sensory score and temperature data, and bar charts of TVC and bacterial counts of five strains. The sensory score pie chart displays the number of samples for a given sensory score value. The sectors are colored green for the lowest sensory score value and red for the highest value. The colors of the values between the highest and the lowest value are taken from the green to red gradient. This coloring scheme is used for the PCA and HCA tools as well. This part of the tool also presents the TVC and bacterial counts in bar charts.
The Principal Component Analysis tool
The Principal Component Analysis tool is used to perform Principal Component Analysis of the selected dataset (either HPLC, FT-IR or eNose data set). The PCA may help in reducing the number of variables in the dataset and in finding any structures in the relations between variables. In meat spoilage research, the PCA may provide a quick exploratory look at the data.
The main functionality of the PCA tool includes: (i) Generation of the dynamic and interactive PCA plot which can be zoomed in, zoomed out and panned using control buttons (ii) A tool for assigning colors to the sample points according to the values of the selected metadata data set. This functionality can be accessed using the Assign colours button (iii) A tool for including or excluding variables from the PCA calculations. This functionality can be accessed using the Exclude variables button (iv) Two scaling methods: Range-scaling and Auto-scaling which can be accessed using the Scaling method menu (v) The Sample names on/off button used to hide or show sample names on the plot (vi) The Biplot on/off button used to hide or show loadings on the plot (vii) The Reset button used to set the PCA plot to the default view (viii) A popup window with additional metadata information when hovering a mouse pointer over a sample point.
Mathematical modeling and Validation of mathematical/multivariate models to predict meat (beef or pork) quality / spoilage
In the previous section it was evident that a great number of multifactorial experiments were designed and performed during the project period. It was also evident that most instrumental and /or analytical techniques were exchanged among the project partners. This included a Raman Spectrograph supplied by the University of Manchester to the Agricultural University of Athens and a Videometer device supplied by Videometer to AUA, while personnel was exchanged between AUA and the University of Manchester, AUA and Naple, Videometer and AUA, and Cranfield and AUA.
In order to achieve the validation studies, several models were developed based on various analytical techniques applied within WP1 and WP3 such as E-nose, HPLC, FTIR, and videometer. Each of the developed models was validated against a testing dataset and the model performance was assessed based on how accurately the model can classify the freshness of a given sample compared to sensorial evaluation and microbiological enumeration. Models evaluation report was generated which help to identify the most suitable modeling approach for a particular analytical platform, as well as the possibility to improve the model performance by developing prediction models based on a combined output from two or more experimental platforms.
Different models have been developed in this project; two indicative cases are reported below;
In the 1st case (Panagou et al. 2011); A series of partial least squares (PLS) models were employed to correlate spectral data from FTIR analysis with beef fillet spoilage during aerobic storage at different temperatures (0, 5, 10, 15, and 20oC) using the dataset presented by Argyri et al. (2010). The performance of the PLS models was compared with a three- layer feed-forward artificial neural network (ANN) developed using the same dataset. FTIR spectra were collected from the surface of meat samples in parallel with microbiological analyses to enumerate total viable counts. Sensory evaluation was based on a three-point hedonic scale classifying meat samples as fresh, semi-fresh, and spoiled. The purpose of the modeling approach employed in this work was to classify beef samples in the respective quality class as well as to predict their total viable counts directly from FTIR spectra. The results obtained demonstrated that both approaches showed good performance in discriminating meat samples in one of the three predefined sensory classes. The PLS classification models showed performances ranging from 72.0 to 98.2% using the training dataset, and from 63.1 to 94.7% using independent testing dataset. The ANN classification model performed equally well in discriminating meat samples, with correct classification rates from 98.2 to 100% and 63.1 to 73.7% in the train and test sessions, respectively. PLS and ANN approaches were also applied to create models for the prediction of microbial counts. The performance of these was based on graphical plots and statistical indices (bias factor, accuracy factor, root mean square error). Furthermore, results demonstrated reasonably good correlation of total viable counts on meat surface with FTIR spectral data with PLS models presenting better performance indices compared to ANN.
In the 2nd case (Argyri et al. 2010); A machine learning strategy in the form of a multilayer perceptron (MLP) neural network was employed to correlate Fourier transform infrared (FTIR) spectral data with beef spoilage during aerobic storage at chill and abuse temperatures. Fresh beef fillets were packaged under aerobic conditions and left to spoil at 0, 5, 10, 15, and 20C for up to 350h. FTIR spectra were collected directly from the surface of meat samples, whereas total viable counts of bacteria were obtained with standard plating methods. Sensory evaluation was performed during storage and samples were attributed into three quality classes namely fresh, semi-fresh, and spoiled. A neural network was designed to classify beef samples to one of the three quality classes based on the biochemical profile provided by the FTIR spectra, and in parallel to predict the microbial load (as total viable counts) on meat surface. The results obtained demonstrated that the developed neural network was able to classify with high accuracy the beef samples in the corresponding quality class using their FTIR spectra. The network was able to classify correctly 22 out of 24 fresh samples (91.7%), 32 out of 34 spoiled samples (94.1%), and 13 out of 16 semi-fresh samples (81.2%).
In general during this project fresh meat either beef or pork was stored under aerobic and modified atmospheres with or without the addition of volatiles compounds from essential oils; these samples were stored in a wide range of temperatures from 0 to 20o C. During storage, packages were subjected to microbiological analyses for the determination
(i)of Total viable counts, pseudomonads, Enterobacteriaceae, yeasts, Brochothrix thermosphacta and lactic acid bacteria.
(ii)to sensory analysis by a four-member taste panel for the characterisation of meat samples in three pre-defined quality classes, namely Fresh (F), Semi-fresh (SF), and Spoiled (S).
(iii)(date were collected form instruments of vibrational spectroscopy e.g. FTIR, RAMAN, fluorescent spectroscopy, multispectral images and electronic nose (iv) metabolic data were collected from the sample using HPLC and GCMS instruments.
Validation of beef (minced) Quality-using FTIR
The classification table resulting from the FDA provided 100 % correct classification and 76.3% correct classification when cross-validated. Even though 23.7% of samples could not be cross-validated, no fresh (F) sample was reclassified as spoiled (S) or vice-versa. The same approach was followed taken the type of packaging as a factor. PCs resulting from the PCA were subjected to a FDA based on the defined packaging type (air, MAP, and active packaging) constituting the dependent variable. The classification table resulting from the FDA provided 100% correct classification and 92.5% correct classification when cross-validated.
(b) Validation of beef (fillets) Quality-using FTIR
Sensory evaluation was performed during storage and samples were attributed into three quality classes namely fresh (F), semi-fresh (SF), and spoiled (S). A neural network was designed to classify beef samples to one of the three quality classes based on the biochemical profile provided by the FTIR spectra, and in parallel to predict the microbial load (as total viable counts, TVC) on meat surface. The network consisted of an input layer with seven input nodes for storage temperature, sampling time, and the five principal components (PCs). The output layer consisted of two nodes, one for the quality class (Fresh, Semi-fresh, Spoiled), and another for the predicted total viable counts of the beef fillet sample. The classification accuracy of the neural network was determined by the number of correctly classified samples in each sensory class divided by the total number of samples in the class. The validation of the neural network in the prediction of TVC for each meat sample analyzed was determined by the bias (Bf) and accuracy (Af) factors, the mean relative percentage residual (MRPE) and the mean absolute percentage residual (MAPR), and finally by the root mean squared error (RMSE) and the standard error of prediction (SEP).
The performance of the MLP network to predict total viable counts in meat samples in terms of statistical indices is presented in Table 3 (Appendix II). Based on the calculated values of the bias factor (Bf) it can be inferred that the network under-estimated TVC in semi-fresh and spoiled samples (Bf less than 1), whereas for fresh samples over-estimation of microbial population was evident (Bf greater than 1). In addition, the values of the accuracy factor (Af) indicated that the predicted total viable counts were 18.1%, 12.2%, and 8.4% different (either above or below) from the observed values for fresh, semi-fresh, and spoiled meat samples, respectively. The standard error of prediction (SEP) index is a relative typical deviation of the mean prediction values and expresses the expected average error associated with future predictions. The lower the value of this index is, the better the capability of the network to predict microbial counts in new meat samples. The value of the index was less than 10% in spoiled samples indicating good performance of the network for microbial count predictions in this class.
(c) Validation of beef (minced) Quality using FTIR and RAMAN
(d) Validation of beef (minced) Quality using multispectral image analysis
Mathematical model development included initially principal components analysis (PCA) for dimensionality reduction of the data set. Subsequently, for qualitative analysis, principal components (PCs) significantly contributing to the variance of the data set were subjected to factorial discriminant analysis (FDA) to predict the class membership of a sample belonging to a previously-defined qualitative group. Finally, for quantitative analysis, PCs significantly contributing to the variance of the data set were regressed using a partial least squares regression (PLS-R) onto the different microbial groups enumerated during storage data in order to predict the microbial load of meat samples directly from spectral data. The first PCA showed that all the bands were found to be significant and thus all of them were used for further analyses. The FDA provided classifications of the samples with an overall correct classification for the validation sensory scores of 71.15%, whilst the provided estimations from the PLS-R model were better regarding the overall classification (86.54%) and each sensory group separately.
(e) Validation of pork (minced) Quality using multispectral image analysis
The quality of stored minced pork meat was monitored using a rapid multispectral imaging device to quantify the degree of spoilage. Bacterial counts of a total of 155 meat samples stored for up to 580 hours have been measured using conventional laboratory methods. Meat samples were maintained under two different storage conditions - aerobic and modified atmosphere packages as well as under different temperatures.
The multispectral imaging device recorded spectra in the visible and the start of the near-infrared area, and these images can be analysed in conjunction with various machine learning and vision techniques. Features were extracted to evaluate the spoilage process of the meat by predicting Total Viable Counts as well as classifying meat pieces into one of three classes, namely fresh, semi-fresh or spoiled, with ground truth being set by a sensory panel. For the multispectral images, an overall classification performance of 76.1% was achieved. For the microbial counts an overall classification performance of 80.0% was achieved. Thus, considering the fact that the electromagnetic area was sampled in only 18 distinct areas, mainly in the visible region, a classification error of 76.1% is a relatively good performance. A very good discrimination between spoiled and fresh pieces of meat was achieved, while semi-fresh meat caused some misclassification, an issue that can be limited if a significantly larger amount of samples could be introduced for analysis. Concluding, in contrast with the retrospective and laborious microbiological analysis, the rapid non-invasive equipment that has been used, requiring no sample preparation, and in which additional parameters have been included in this work such as time, temperature and atmosphere provided a rather promising tool for assessing pork spoilage.
(f) Validation of pork (minced) Quality using FTIR spectral data
A three-class evaluation scheme was employed in this experiment. The first class (Fresh) corresponded to the absence of off-flavors, equal to the reference sample; the second class (Semi-fresh) corresponded to the presence of slight off-flavors but not spoiled (still acceptable quality); and the third class (Spoiled) corresponded to clearly off-flavors development (unacceptable quality). Semi-fresh was the first indication of meat spoilage (incipient spoilage) in which the sample was marginally accepted. For qualitative analysis, PLS discriminant analysis (PLS-DA) was used to develop models allowing the discrimination of meat samples in the selected sensory classes. Good classification was obtained in all sensory classes demonstrating the effectiveness of the method as a rapid screening technique to monitor minced pork spoilage. Specifically, a number of 6 LVs was finally selected as input variables in PLS-DA model development, presenting the highest correct classification (%) in the training (94.0%) and test (88.1%) datasets.
For the training dataset, the PLS-DA approach provided 100% correct classification for spoiled samples, whereas for fresh and semi-fresh the respective number was 93.3% and 86.7%, respectively, representing 1 misclassification out of 15 fresh samples and 2 misclassifications out of 15 semi-fresh samples. PLS regression (PLS-R) models were built for the counts of TVC, Pseudomonas spp., B. thermosphacta and lactic acid bacteria using FTIR responses as input variables and the microbial counts as output variables. Good relationships were found between the results of FTIR and the microbiological analysis.
(g) Validation of pork (minced) Quality using electronic nose (e-nose).
In this part of the study, e-nose was used to obtain volatile fingerprints of minced pork meat during storage in different conditions (i.e. temperature and packaging systems) in an attempt to monitor spoilage. The eight sensors had different responses to the samples tested and this could be attributed to the intrinsic selectivity of the molecular sensing mechanism and to the mass of the molecules that are bounded at the coated surface of the sensors. The volatile patterns collected from the eight sensors signal of e-nose were subjected to Principal Component Analysis (PCA) in order to reduce the dimensionality of the data set. The total variance (100%) of the data set could be explained by eight principal components (PCs) among which the first five accounted for 99% of total variance observed in the experiment.
(h) Validation of beef (fillets) Quality using electronic nose (e-nose).
The volatile patterns collected from e-nose were initially subjected to Principal Component Analysis (PCA) to reduce multi-collinearity (e.g. sensors with overlapping sensitivities) and allow the information to be displayed in a smaller dimension. Subsequently, the scores of the first five principal components accounting for 99% of total variance observed in the experiment were further used as input in Support Vector Machines (SVM) analysis using linear, polynomial and radial basis function (RBF) kernels, to predict the quality of a meat sample that was pre-characterized as fresh (F), semi-fresh (SF) or spoiled (S) from a taste panel. The approach was similar to this mentioned for validation of pork model. Test data were not employed in any step of model development, but they were used exclusively to determine its performance. The classification accuracy of the SVM model was determined as the number of correctly classified samples in each sensory class divided by the total number of samples in the class. Results showed that SVM with RBF kernels provided good discrimination of beef fillet samples regarding their spoilage status at both packaging conditions.
(i) Validation of beef (fillets) Quality using ensemble-based systems and electronic nose (e-nose).
The bagging approach was followed to develop an ensemble-based system for predicting quality based on the sensory evolution scores given by the panel as described earlier. The dataset was split into training and a testing subset. For each of the classifiers forming the ensemble, the training subset was boostrapped to form a subset of 330 samples, which is then divided further into training and a testing subsets in order to perform the grid search optimization process. A total of 200 classifiers were developed to form the ensemble. The ensemble prediction accuracy was calculated using the testing subset and the final output was computed using various aggregation methods. As shown in Table 15 (Appendix II), the bagging approach has improved the overall prediction accuracy by more than 10% when compared to the single classifier performance when assessed using an unseen testing subset, showing a performance of approximately 83%. All aggregation methods applied for output fusing performed equally well. Naive Bayes aggregation showed the best classification performance, with an overall accuracy of 84.10%.
Potential Impact:
Introduction
Economic Impact on Meat industry - a EU Need
The European Union Beef Market; The Economic Impact of the Meat Products Industry begins with an accounting of the direct employment in the various sectors. Meat production encompasses slaughterhouses, packers, company- owned distribution and supply operations and importers. Wholesaling includes the nationwide network of meat and meat product wholesalers and related warehouse and transportation units. Retailing includes locations where meat is consumed on-premise, such as restaurants. Off-premise retail outlets are supermarkets, butchers, warehouse stores, and similar locations. The data comes from a variety of government and private sources. It is sometimes mistakenly thought that initial spending accounts for all of the impact of an economic activity or a product. For example, at first glance it may appear that consumer expenditures for a product are the total sum of the impact on the local economy. However, one economic activity always leads to a ripple effect whereby other sectors and industries benefit from this initial spending. This inter-industry effect of an economic activity can be assessed using multipliers from regional input-output modeling. The economic activities of events are linked to other industries in the state and national economies. The activities required to produce a can of packaged meat from butchering a hog, to packaging, to shipping generate the direct effects on the economy.
Development of the EU beef industry; in order that the beef industry remains strong in a market-oriented environment it was reported that beef quality specifications, in the whole meat chain, will again become the primary market driver, both from a consumer and producer perspective1. For those EU beef value-chain participants that are able and willing to adapt to the new challenges, the future could offer significant opportunities, but still, the way ahead remains essentially uncharted territory with many cross road decision points. The EU will continue to be a dominant beef consumer and domestic production will need to be at substantial levels to meet this demand.
General Impact of SYMBIOSIS-EU project
The Symbiosis project, has a significant impact on the following key communities within the EU: (i) European consumers of meat products, who will benefit from better quality products, with higher levels of food safety (ii) The European food industry, which will be able to deliver improved products, whose quality and safety can be proven scientifically (iii) The food microbiology community, who will gain new tools and new insights into the mechanisms of food spoilage (iv) The research community in general, who will benefit from new software tools for data integration, and a high profile real world example of their technologies in action.
The project contributed to the general policy of the EC to improve the quality of health and safety delivered to the European citizen regarding the food chain. The safety of food products is currently very much in focus, owing to calamities with microbiological outbreaks, dioxin contamination and other threats to human health (https://webgate.ec.europa.eu/rasffwindow/portal/index.cfm?event=notificationDetailandNOTIF_ REFERENCE=2011.0805 and http://www.foodpoisonjournal.com/foodborne-illness-outbreaks/7-children-in-france-remain-in-the-hospital-due-to-e-coli-infection/).
SPECIFIC IMPACTS
Impact for the industry:
Development of simple, effective and inexpensive evaluation of the meat quality and safety sector (WP 4, and WP5);
Development of active packaging systems for preservation of meat with which safety and quality was assured (WP1);
Development of new vibrational spectroscopies, surface chemistry, fluorescent spectroscopy and electrical nose, for the simple and rapid detection of microbial metabolites indicative of specific muscle quality and safety attributes (WP2 and WP3);
Development of advanced diagnostic methods based on intelligent and statistical schemes to evaluate meat quality (WP2-5, WP7-8);
Development of mathematical models to predict shelf life as well as formation of spoilage compounds (WP5 and WP7).
b- Impact for the society:
The website with current information on the importance of microbial metabolites in meat and other project related findings and information for the dissemination to non-specialist laboratories and Small Medium Enterprises (SMEs) (WP2 and WP8);
Develop and distribute user-friendly application software that facilitate practical use of the methods and mathematical models that have been successfully validated in this project, and thereby provide tools that can improve the competitiveness of the European muscle food sector (WP4 and WP5);
The availability of highly effective sensors have broaden the market opportunities for companies involved in manufacture of electronic systems and components, thus having a positive impact on employment within this sector. In addition new vacancies for qualified engineers to use such automated systems will be available in the food industrial sector (WP5, WP7 and WP9).
c- Impact for the science
Provide basic knowledge tools and resources for the interpretation and management of the generated data (e.g. genomic, proteomic, metabolomic and phenotypic data) relevant to microbial behavior in relation to safety and spoilage of food (WP1, WP2 and WP6);
Provide fundamental knowledge (genomics, metabolomics) of the freshness, spoilage and hazards (e.g. biogenic amines, etc) associated with meat products stored under dynamic conditions, enabling the rational design of procedures for reduction of food losses and to increase quality, consumer satisfaction and public health protection (WP2 , WP6 and WP7).
Impact of Specific Work Packages
Impact of WP1: Monitoring meat microbial quality and safety during production, storage and distribution.
The first task toward an integrated understanding of the Food microbial ecology (the meat ecosystem) was to identify and study the most important components. The impact of this wp was very important. It was shown that storage temperature combined with packaging conditions as well as the application of different treatments/antimicrobials induced a selection of the spoilage microbiota even at strain level.
Impact of WP2: Monitoring of major profiling of metabolites (metabolomics) occurring in meat during production, storage and distribution.
This work package plays a major role in offering a better understanding of the spoilage process within meat products. This involves a series of microbiological assessments of meat quality, in order to identify major metabolic profiling of metabolites throughout the complete procedure of production, storage and distribution.
Impact of WP3: Development and application of an image / sensor-based platform to monitor meat quality.
The conventional microbiological analysis of food products usually focuses on ensuring the arrival of safe product to the customer. On the other hand, from the customer point of view, the product should be satisfactory in terms of appearance, odour and taste. One important task within WPs 2 and 3 is the in-depth study of the chemical changes occurring in meat products under different storage conditions responsible of these sensorial changes. This would have a significant impact on the food industry, as it will provide a more detailed scientific understanding of the effects of different packaging solutions, enabling packaging to be optimised according to the needs of the packed product and the supply chain.
Impact of WP4: System integration
The impact of Integration of the prototype with the Symbiosis Online Research Framework SORF, was indeed priceless since it was doable to combine and upload (incorporate) data (heterogeneous) complete experimental dataset files instead of having to manually add datasets entities to the interface.
Impact of WP5: Mathematical modeling of metabolite formation and sensor responses
This work package, which deals with the mathematical analysis of the data collected in earlier work packages, has a profound impact on the food microbiology sector, both in terms of the discovery of new biological knowledge of interest to industry and academia, and also in terms of demonstrating to the community the benefits of multivariate data analysis in their field. Indeed the traditional conventional analysis of data for microbiologists (and biologists in general) has taken the univariate approach to interpreting their data e.g. looking at a single marker or characteristic at a time. Given the much more powerful analytical approaches now available (and used in this project) multivariate techniques are essential if maximum information is to be extracted from the data.
Impact of WP6: Understanding regulation of metabolomics for meat quality.
A full understanding of metabolic pathways in the spoilage dynamic would be practically impossible without the development of genomic-scale metabolic network. This approach relies on the assembling of various data sources pertaining to all the chemical reactions that take place during the spoilage process. Microbiological and chemical data represent the strongest evidence to the occurrence of a metabolic reaction (through the identification of biomarkers relating to this reaction); this is often referred to as the top-down approach.
Impact of WP7: European validation studies with indices of quality and mathematical models from individual countries.
(i) At industrial level: Promising techniques developed in WP4, and successfully applied through WP1, have tested in-site. The ultimate goal of this task was to quantitatively assess the accuracy of applying these methods in a practical, realistic environment.
(ii) At laboratory level: One of the major challenges facing the practical application of results obtained through meat science research is the lack of agreement between data obtained in different countries/laboratories. In this project the advantage of the availability of broad range of scientists from different laboratories in order to provide a better common understanding of microbial metabolite indices has been applied. This would take place through the cross-validation of results obtained from one laboratory by other beneficiaries' labs.
(iii) At computational level: Knowledge gained from the previous two tasks would serve as a refining loop for the mathematical models in order to develop sufficiently accurate models able to lead new biological experiments, and predict dynamic conditions that were not identifiable through traditional methods.
Impact of WP8: Dissemination including development and distribution of application software.
Making the project outputs we have maximized the impact of the project. Indeed the following data are available among the partners
(i)Acquired data: All analytical data collected during the project will be made available to the community via the data integration system developed in WP4.
(ii)Research findings: Information derived from the acquired data using the mathematical and other data analysis techniques employed will be reported though oral presentations at conferences and in peer reviewed journal articles.
(iii)Software developed: The software underpinning the data integration system will be provided to others wishing to set up a data integration system for their own projects. Similarly, software tools implementing the most effective data analysis techniques identified in WP5 will be made available via the project's web page to allow people to carry out their own analysis without special software or skills.
Results transferred to market and to industry
The main results of this research are expected to improve the competitiveness of the European meat industry. For this reason the project has provided a source of information and education of industrial personnel, on the potential of harmonious coexistence of science and new technology, quality and safety assurance of these highly perishable products. Training has been already performed and involved placement of academic staff within the food industry and the placement of industrial staff in academic institutions. This has enabled both sides (researchers and members of the meat industries) to be trained in the requirements of the meat industry and understand the potential for incorporation of these new devices and tools (e.g. multivariate mathematical modeling approaches) into their business.
List of Websites:
http://www.symbiosis-eu.net
http://www.symbiosis-eu.net/
Contact (coordinator) person; George-John NYCHAS gjn@aua.gr or nychas@hol.gr
The SYMBIOSIS-EU is a European project that brings together 14 beneficiaries from 6 EU countries (plus one each from NZ and US) to study tools that will allow a rapid evaluation of meat quality and safety in comparison with the time consuming microbiological analyses or the unattractive and expensive sensory analysis, of which both are most often used to evaluate the freshness, spoilage and/or safety of meat and meat products (EU 2005).
The overall aim is to identify and quantitatively evaluate practical and easy to use chemical, biochemical and molecular indices and establish their applicability as quality monitors for inspection of meat safety and quality. The project has applied a multidisciplinary system wide approach relying on converging technologies (bioinformatics, nanotechnology, modeling) to obtain knowledge for meat quality that has been translated into simple devices and practical indicators of quality and safety.
The main objectives met in this project were (i) to develop and/or validate easy to use chemical/biochemical methods (e.g. biosensors, fluorescence, FT-IR), molecular methods (DNA, microarrays), (ii) to develop a suitable software platform for data sharing and integration, (iii) to apply multivariate statistical methods and machine learning (neural networks, fuzzy logic, genetic algorithms) to identify robust multiple compound quality indices, (iv) integration of the sensors and information platform and development of a system to automatically transform data acquired from a sample into a diagnosis of meat safety and quality.
Project Context and Objectives:
Project Description
The current practice to assure the safety/quality of meat still relies heavily on regulatory inspection and sampling regimes. This approach, however, seems inadequate because it cannot sufficiently guarantee consumer protection since 100% inspection and sampling is technically, financially and logistically impossible. In addition, the following issues need to be considered: (i) more than 50 chemical, physical and microbiological methods have been proposed for the detection and measurement of bacterial safety or spoilage in meat, (ii) most of these methods are time-consuming and provide retrospective information and thus they cannot be used on- or at-line, (iii) the lack of general agreement on the early quality changes, and (iv) the changes and development in technologies for food processing and preservation [e.g. vacuum packaging (VP), modified atmosphere packaging (MAP), active packaging, etc] make it evident that the important and urgent task of identifying safety and spoilage indicators is a complicated proposition.
Beyond the Baseline data
The basic concept in this project is related with the integration of the molecular tools (WP1, WP6) as well as sensors /biosensors (WP3) and other technologies (e.g. FT-IR ATR, RAMAN) with the information platform (WP4) and to develop an expert system (M4.1) able to automatically classify the sensorial input into a diagnosis based on extracted pre-processing features (WP4 and WP5). Additionally this project will generate data covering genomic, proteomic, metabolomic, and phenotypic aspects. This, therefore, requires the integration (WP4 and WP5) of highly divergent data generated from the various high throughput analytical techniques such as HPLC, GC-MS, Microarray, Electronic nose, Fluorescence Detectors and multi-spectral imaging (M4.1 M4.3). Due to the heterogeneous nature of the data generated, and the geographic distribution of the project beneficiaries, an important task will be to create a central data (M4.2) repository that will serve as a common framework for the research project.
Specific Objectives per WorkPackage
Main objectives of WP0
The objectives for the reporting period regarding WP0 were: (i) The Overall Project Management (Task 0.1) (ii) To fulfil the general coordinators responsibilities (Task 0.2) (iii) Collate the beneficiaries financial and technical contributions and submit reports to the EU Commission. In the meetings the GA, and GB were also organized (Task 03; Subtask 0.3.1 and Subtask 0.3.2 was achieved last June in Brussels). A revised plan was delivered after the review and the assessment report for the reviewers.
Main objectives of WP1
This WP was finalized in month 30; the main objectives related to the reporting period were: (i) To finalized the identify and characterize ESO associated with the spoilage of meats produced and stored under different temperatures and packaging systems (MAP/VP). (ii) To collect information on the spoilage process of meat that will lead to reliable and specific kinetic spoilage models; both these Tasks 1.1 and task1.2 completed well.
The characterization of microbial association of various meat and meat products has been completed and most of the data have been published or have been submitted for publication.
Main objectives of WP2
Similarly this WP was ended in month 30; during the last 18M period the main objective was: To determine the effect of intrinsic parameters (physical, chemical and structural properties inherent in meat itself) and extrinsic parameters, i.e. temperature imposed atmosphere, microbial association, quorum sensing, on the formation of microbial metabolites by ESO. This was achieved and the corresponding deliverable was provided on time. Most of the results have been published or have been submitted for publication. Additionally this database was subject to WP5, WP7 and has been used successfully to fulfill the achievements of the above mentioned WPs.
Main objectives of WP3
The WP ended in month 30; in the running period the main objective of wp3 was to record all the derived data which should have been used to develop models / tools that have been proved to be able to determine the potential for spoilage of meat products. An additional output will be the information about the potential for such approaches to detect the metabolites of food-borne pathogenic bacteria. The huge data set derived from this WP, includes spectra from vibrational spectroscopy instruments, e.g. FTIR, Raman, surface chemistry as measured with VIDEOMETER, as well as the metabolomics, e.g. data from HPLC, GC/MS and e-nose, were the basis to feed the platform developed by other beneficiary for WP4, WP5 and WP7.
Main objectives of WP4
In the running period the main objectives of wp4 were: (i) To develop a database by capturing all of the experimental data from the project. The database will be specifically designed to cope with the heterogeneous nature of the sensorial data acquired during the project and (ii) To develop an integrated end-user interface to visualize all acquired information/results.
Main objectives of WP5
The main objectives of wp5 were: (i) The evaluation of the sensorial information through on reliability of information, redundancy, and discriminative character; and (ii) The development of a diagnostic system to classify automatically the sensorial input into a decision based on the extracted features.
At the end of the project two more objectives should be completed: (i) The development of modeling tools for meat spoilage, and (ii) Kinetic and prediction modeling.
Main objectives of WP6
The main objective of this wp for the first 18M was: To investigate the expression of key genes involved in synthesis of specific spoilage metabolites under different environmental conditions induced by storage and packaging. The 2nd objective is to investigate the expression of virulence and stress response related genes under different environmental conditions induced by storage and packaging.
Main objectives of WP7
The aims of WP 7 were to evaluate indices of quality and mathematical models for metabolite formation at a European level. In particular, an evaluation of differences between data from laboratories in separate countries will be possible because common analytical methods will be used within the present project (Task 7.1 and Task 7.2). It should be noted that several beneficiaries of the project possess equivalent advanced equipment like GC-MS and HPLC instruments or transportable miniaturized sensors (Task 7.2) that may easily allow on-line data recording in plants in several EU countries. Data from Task 7.1 and 7.2 will be used for validation of mathematical models developed within the project in WP4 and WP5. The produced models will be evaluated at European level (Task 7.3).
Testing the developed techniques, tools and methodologies, validation of the mathematical models in real time (D7.1 D7.3) (M30, M36), will give the opportunity to measure the efficiency of the produced models/tools. The verification of this models/tools (D7.2; M30) can be used to elaborated further the meat spoilage.
Main objectives of WP8
In this WP the last period the main objectives are: (i) To include successfully validated models in application software and thereby facilitate their practical usefulness, and (ii) To establish permanent Training and Dissemination Units at various Meat and other Industries in selected EU countries. Since the development of SORF was proved to be a general tool, this exhibition was done to different industries also.
Main objectives of WP9
The main objective of this wp for the project period was: Communication and transfer knowledge from all WP (Task 9.1). Organisation of training sessions and implementation of distance learning packages (e.g. bioinformatics, molecular tools, chemometrics, mathematical modeling, etc). Training activities (Task 9.1) will deliver both traditional (e.g. residential courses; within Universities, company training days, plant visits; and web-based (Task 9.2) as well as in workshops and conferences.
Project Results:
Effect of Microbial Association on meat Spoilage
Meat is recognized as one of the most perishable foods. Apart from the physical damage, oxidation and color change, the other spoilage symptoms are due to the undesirable growth of microorganisms to unacceptable levels. The microbial quality of meat depends on the physiological status of the animal at slaughter, the spread of contamination during slaughter and processing, the temperature and other conditions of storage during distribution (Nychas et al. 2008). When large numbers of undesirable microorganisms are present in raw meat, then there will be undesirable changes such that the food becomes unappealing and unsuitable for human consumption (Nychas et al. 2008). Many groups of organisms contain members potentially contributing to meat spoilage under appropriate conditions. This makes the microbial ecology of spoiling raw meat very complex and thus the spoilage very difficult to prevent.
Development of microbial association
Within the SYMBIOSIS EU project attention has been also given to the characterization of the spoilage microbiota not only at species but also at strain level. This is in fact believed to potentially play a pivotal role in the establishment of the meat spoilage as different strains of the same species may have different spoilage activities or can be differently affected by antimicrobial storage conditions (Andritsos et al. 2012; Casaburi et al., 2011; Doulgeraki et al. 2011; Ercolini et al., 2010b). Therefore, characterization of lactic acid bacteria (Doulgeraki et al. 2010), Enterobacteriaceae, Br. thermosphacta, Pseudomonas spp., and Carnobacterium spp. (Casaburi et al., 2011; Doulgeraki et al. 2011; Ercolini et al. 2010b; Doulgeraki and Nychas 2012 ; Papadopoulou et al. 2012) isolated during storage of meat under different conditions (aerobic, vp, MAP with or without essential oils) have been recently performed. Conventional and molecular tools that have been used in this project for detection and identification of the bacteria are also presented and discussed in a recent publication (Doulgeraki et al. 2012).
Succession of specific groups
With regard to Lactic Acid Bacteria (LAB), a total of 266 strains have been isolated from minced beef stored at 0, 5, 10 and 15o C aerobically and under modified atmosphere packaging consisting of 40% CO2-30% O2-30% N2 in the presence MAP (+) and absence MAP (-) of oregano essential oil. Sequencing of their 16S rRNA gene along with presence of the katA gene demonstrated dominance of the LAB microbiota by Leuconostoc spp. during aerobic storage at 5, 10 and 15o C as well as during MAP (-) and MAP (+) storage at 10 and 15o C Lactobacillus sakei prevailed during aerobic storage at 0 as well as at MAP (-) and MAP (+) storage at 0 and 5o C. The sporadic presence of other species such as Leuconostoc mesenteroides, Weisella viridescens, Lactobacillus casei and Lactobacillus curvatus has also been determined. Pulsed-Field Gel Electrophoresis of high molecular weight genomic DNA revealed the dynamics of the isolated LAB strains. Prevalence of Leuconostoc spp. was attributed to one strain only. On the other hand, packaging conditions affected Lb. sakei strain spoilage dynamics.
Effect of natural inhibitors on pathogenic bacteria
Several studies (Argyri et al. 2010; Ferrocino et al 2012; La Storiaet et al. 2008) in this project have demonstrated the inhibitory effect of volatiles from Oregano Essential Oil against the microbial association of meat either beef or pork. The results have been mentioned above. Additional studies have reported the effect of 3 bacteriocins and this was exploited with the atomic force microscopy (AFM) analysis for the characterization of plastic films activated with antimicrobial agents. The aim was to acquire information on the distribution of the antimicrobials on plastic materials with the ultimate scope of understanding the mechanisms of interaction between antimicrobials and materials to be used for food packaging. Four polyethylene films differing in linear, EVA, and erucamide content were activated by 3 different bacteriocins as antimicrobials, namely, nisin and bacteriocins Bac162W from Lactobacillus curvatus and BacAM09 from Lactobacillus plantarum. The spectrum of activity of the bacteriocins was assayed and shown to include several strains of Listeria monocytogenes. The plastic films were activated by a previously developed coating procedure and the surfaces of the active films were examined by AFM. In addition, roughness parameters related to the single surfaces were investigated by appropriate software.
Contribution of Energy Sources and microbial metabolites on meat spoilage
It is well established that low molecular weight compounds, e.g. glucose, lactic acid and certain amino acids are catabolized by almost all the bacteria of the meat microflora, followed by nucleotides, urea and water-soluble proteins. The former compounds are the essential energy sources for the massive growth of the microcosm on the meat despite their negligible quantity in comparison to proteins. It is shown that actual concentrations of these compounds can affect the type (e.g. saccharolytic, proteolytic) and the rate of spoilage, and moreover they seem to be the principal precursor(s) of those microbial metabolite(s) that we perceive as spoilage (Nychas et al. 2008).
There are three classes of substances that are used by the microbial association: (i) compounds involved in the glycolytic pathway (e.g. glycogen, glucose, lactate, etc.), (ii) metabolic products (e.g. gluconate, pyruvate, lactate, etc), and (iii) nitrogen energy sources (e.g. aminoacids, proteins). Glucose has been found to be the precursor of many off-odours during meat storage (Nychas et al. 2008), while its limitation could cause a switch from a saccharolytic to an amino acid degrading metabolism in at least some bacterial species. The changes of glucose and lactate as well as their oxidative products (e.g. gluconate) have been proposed for use in describing or predicting the degree of spoilage (Nychas et al. 2008). The information, about the rate and the sequence of utilization of available substrates in space and time, by the ESO in meat stored under different (dynamic) conditions, is rather negligible. The nutrient contribution may be related to positive (synergistic/syntrophic) or competitive activities for nutrients/energy (e.g. under excess, limitation or starvation), while metabiosis (production of a favorable environment) and cell-to-cell communication (quorum sensing) could also affect the physiological attributes of the organisms under the imposed ecological determinants.
In a specific study, beef chops were stored at 4°C in different conditions: air (A); modified atmosphere packaging (MAP); vacuum (V); bacteriocin-activated antimicrobial packaging (AV). After 0 to 45 days of storage, analyses were performed to determine loads of spoilage microorganisms, microbial metabolites by SPME-GC/MS and 1HNMR, microbial diversity by PCR-DGGE and pyrosequencing.
The metabolic profile of minced beef stored (i) aerobically, (ii) under modified atmosphere packaging (MAP), and (iii) under MAP with oregano essential oil (MAP/OEO) at 0, 5, 10, and 15 °C was investigated suing HPLC and GC/MS (Argyri et al. 2011,2011c). The analysis of chromatograms from HPLC resulted in the selection of 17 discrete peaks (with purity greater than 99%). Fifteen (15) of them are presented in all samples whilst the peaks with RTs of 20.5 and 28.1 min were evident only when spoilage was very pronounced. From the metabolomics point of view, the information contained in all peaks is critical for the characterization of spoilage. In this respect, the whole profile (all the peaks) as derived from spectral data was taken into consideration. Since some of the peaks were not identified and thus concentration values could not be calculated for these peaks, the analysis was continued using the chromatographic area of the peaks. The results demonstrated changes in the chromatographic areas under the peaks of the eluted acids, that were associated or not with storage conditions (e.g. temperature, packaging).
Development and application of vibrational spectroscopy and image /sensor based platform to monitor meat quality
Since the evaluation of the degree of meat spoilage and/or hygiene level is usually made either subjectively based on qualitative criteria or by time consuming, destructive microbiological analyses that do not provide the â??immediate answer requiredâ?? (Nychas et al. 2008). Some approaches are based on the use of Vibrational Spectroscopy, which has the advantage over infrared since it is non-invasive, non-destructive, and the sampling protocols are simplified. This is because it can simply analysed through glass containers, and the sample may be simply analysed in its raw form without dilution. Instruments of vibrational spectroscopy methods among others are FT-IR, Raman spectroscopy.
On the other hand, electronic noses are instruments used to classify and identify complex odours since they reproduce structure and principles of the natural olfactory sense. Basically the electronic nose is composed of an array of sensors and a data analysis system. Sensors act like biological receptors and data analysis system allows to transpose information that sensors extract from an odor in an olfactory image analogous with our sensation of a smell. The identification and comparison of odors is possible because different odors have different olfactory images. This distinguishes the Electronic Nose from other analytical techniques, e.g. gas chromatography which identifies and measure single molecular compounds inside a gaseous mixture (Papadopoulou et al. 2011a, 2012).
The main findings related to the mentioned techniques are summarised below.
The results showed that electronic nose, FTIR, Raman Spectroscopy, surface chemistry as expressed by the multispectral instrument and fluorescent spectroscopy could offer a powerful and rapid tool (i) for quality evaluation of fresh meat (beef, pork) e.g. organoleptic shelf life, (ii) for the measurement of the population of different microbial groups, and (iii) can be applied for the prediction of the microbial loads of the different microbial groups.
The FTIR spectral data collected, were analysed and correlated with the microbiological data and the sensory scores in order to build qualitative and quantitative models that predict the spoilage status of an unknown meat sample. A number of studies, being the outcome of this Project, have demonstrated the potential use of this non-invasive technique (Argyri et al. 2009, 2010; Brewster and Goodacre 2012, Nicolaou et al. 2011). In these articles, results have been analysed with Principal Components Analysis (PCA) and Factorial Discriminant Analysis (FDA), partial least squares regression (PLS-R) models and Support Vector Machine (SVMs) models. Moreover, different pre-processing methods applied have improved the performance of the models (Xu and Goodacre 2011, Argyri et al. 2012).
The RAMAN spectra were introduced to minced beef which was stored aerobically and under modified atmosphere packaging (40%CO2/ 30%O2/ 30%N2) at 5°C and microbiological analysis were performed in parallel with pH measurements, and sensory analysis. Moreover, minced beef samples stored aerobically and under the same modified atmosphere with and without an oregano essential oil (OEO) (2% v/w) slow release system at four temperatures (0, 5, 10 and 15°C), where analysed with Raman spectroscopy. The data derived from the microbiological analysis (Total Viable Counts, Pseudomonas sp., Brochothrixthermosphacta, lactic acid bacteria (LAB), Salmonella Enteritidis) were highly correlated with the Raman spectra (Argyri et al. 2012).
The portable fluorescence spectrometer device equipped with an excitation LED, a fibre optic probe and a mini-spectrometer was tested to quantify minced beef meat spoilage during storage and provided an excellent prediction accuracy of the regression models obtained for meat stored under vacuum and aerobic atmosphere.
Similar results for prediction of microbial flora on minced beef were obtained from portable spectrofluorometer (Kaddou et al. 2011).
The feasibility of Fourier transform infrared (FT-IR) spectroscopy to quantify biochemical changes occurring in fresh minced pork and beef meat in the attempt to monitor spoilage has been applied in this project. A number of articles have been already published (Papadopoulou et al. 2011b; Panagou et al. 2011). For this reason, partial least squares (PLS) models and ANN were constructed to correlate spectral data from FTIR with minced pork meat spoilage during aerobic storage of meat samples at different storage temperatures (0 to 20°C). Spectral data were collected from the surface of meat (both pork and beef) samples in parallel with microbiological analysis to enumerate the population of total viable counts, Pseudomonas spp., Brochothrix thermosphacta, lactic acid bacteria and Enterobacteriaceae. Qualitative interpretation of spectral data was based on sensory evaluation, using a three point hedonic scale, discriminating meat samples in three quality classes, namely fresh, semi-fresh and spoiled. The purpose of the developed models was to classify minced pork samples in the respective quality class, and also to correlate the population dynamics of the microbial association with FTIR spectra. The obtained results demonstrated good performance in classifying meat samples in one of the three predefined sensory classes. The overall correct classification rate for the three sensory classes was 94.0% and 88.1% during model calibration and validation, respectively. Furthermore, PLS regression models were also employed to provide quantitative estimations of microbial counts during meat storage.
B-E- nose
The performance of a quartz microbalance based electronic nose has been evaluated in monitoring aerobically and under MAP packaged beef fillets spoilage at different storage temperatures (0 to 20°C). Electronic nose data were collected from the headspace of meat samples together with microbiological analysis for the enumeration of the population dynamics of total viable counts, Pseudomonas spp., Brochothrix thermosphacta, lactic acid bacteria and Enterobacteriaceae. Qualitative interpretation of electronic nose data was based on sensory evaluation discriminating samples in three quality classes, i.e. fresh, semi-fresh, and spoiled. Support Vector Machines (SVM) classification and regression models using radial basis kernel function were developed to classify beef fillet samples in the respective quality class, and also to correlate the population dynamics of the microbial association with sensors responses. The obtained results demonstrated good performance in discriminating meat samples in one of the three pre-defined quality classes.
C-Fluorescence Spectrometer
Kaddour et al. (2011) used a portable fluorescence spectrometer to quantify minced beef spoilage. This study was investigated on samples stored aerobically and under vacuum at 5 and 15 °C. Total viable counts (TVC), Pseudomonas, lactic acid bacteria, and yeast/molds counts were investigated with classical culture methods. Fluorescence spectra were recorded on the same samples using different excitation LEDs (280, 320, and 380 nm). PLS-R with leave-one-out cross validation was used to perform calibration and validation models. The PLS-R models presented good R2CV (0.50-0.99) and ratio of standard deviation (RPDCV: 1.40-8.95) values after cross-validation. It could be concluded that portable spectrofluorimeters are promising devices to evaluate spoilage in minced beef.
D-Image analysis
The rapid multispectral imaging device was used to monitor quality and to quantify the degree of spoilage of stored minced pork meat (Dissing et al. 2012). Bacterial counts of a total of 155 meat samples stored for up to 580 hours have been measured using conventional laboratory methods. Meat samples were maintained under two different storage conditions - aerobic and modified atmosphere packages as well as under different temperatures. Besides bacterial counts, a sensory panel has judged the spoilage degree of all meat samples into one of three classes. Results showed that the multispectral imaging device was able to classify 76.13% of the meat samples correctly according to the defined sensory scale. Furthermore, the multispectral camera device was able to predict total viable counts with a standard error of prediction of 7.47%. It is concluded that there is a good possibility that a setup like the one investigated will be successful for the detection of spoilage degree in minced pork meat.
E-Raman Spectroscopy
The results are analysed in the validation section.
Understanding Regulation of Metabolomics for Meat Quality
A number of studies were performed to investigate the expression of key genes involved either in synthesis of specific spoilage metabolites and on virulence and stress response under different environmental conditions induced by storage and packaging. In brief, transcriptomic analyses related to the regulation and control of microbial metabolite profiles of Pseudomonas during meat spoilage under various storage and packaging conditions was performed. In addition, transcriptomic analysis related to genes involved in virulence and stress response in Salmonella during exposure of meat to various storage and packaging was performed.
In the case of Salmonella, the stress response of Salmonella Thompson the exposure of sublethal concentrations of thymol (0.01%) at 37°C was studied (Di Pascua et al. 2010). Emphasis was given in the expression of Thymol, which is a natural biocide and component of some essential oils from herbs. Although it's inhibitory effect on the growth of different microorganisms is well documented, the precise targets of the antibacterial action of thymol is not yet been fully established, the action seems to take place in different ways. The strain Salmonella enterica serovar Thompson MCV1 was grown in the presence of a sublethal concentration (0.01%) of thymol. The proteins extracted from treated and untreated cells were subjected to 2-D PAGE, followed by in-gel spot digestion and subsequent MALDI-TOF analysis. In parallel, control and exposed S. Thompson cells were collected and the gene expression of several virulence genes by qPCR analysis, i.e. hilA, stn and sopB was studied. The analysis of gels showed many proteins that were either upregulated or downregulated by the presence of thymol, with significant changes in proteins belonging to different functional classes. In particular, the thioredoxin-1 was not expressed in the treated cells, indicating that its absence could be a consequence of the stress caused by the presence of thymol. On the other hand, different chaperon proteins were upregulated or de novo synthesis such as GroEL and DnaK, key proteins in the protection mechanism toward thermal stress. Outer membrane proteins were upregulated in treated cells; indeed the bacterial envelope stress response is trigged by the accumulation of misfolded outer membrane proteins.
Experimental dataset file upload system
Before uploading an experiment dataset, a new experiment instance has to be defined within the system. This is done through a Web interface page for defining an experiment name, description, type and data (See Figure 1a). Once these fields are filled, the user can then navigate to the next page where he is provided with a file upload system. The experiment type specified by the user in the first navigation page specified the file upload fields present in this step.
Integration of multi-sensor devices output
The design of the SORF back-end has been extended to integrate the experimental output of many sensorial platforms and experimental data into one experiment entity. One experiment stored with the Symbiosis system can have several dimensions of datasets. These datasets can be coming from heterogeneous experimental platform.
System overall architecture
The design and architecture of the Microstudio platform (Now called The Symbiosis Online Research Framework â?? SORF) has gone through various stages of development, as new technolpogies have been made available throughout the timeline of the Symbiosis-EU project (Mohareb and Bessant 2009, Mohareb 2011a,b). The initial front-end design was implemented using Java Enterprise 6 and JSP, deployed on Sun Application Server 8. This was then replaced at later stages of development by the Java Server Faces 2.0 technology, a more advanced and robust framework that allowed us to provide a more intuitive and interactive interface to the system users.
The Data Layer
The Symbiosis MySQL database is a crucial part of the SORF web application. Its main purpose is the user, experiments, datasets and data management. The Symbiosis Database is composed of two parts. The first part is responsible for storing user information such as name, contact details and organization as well as user's publications. The second part is responsible for storing experiments and their datasets. This work focuses only on the usage of the second part of the Symbiosis Database. The idea behind the Symbiosis Database is that the user may create experiments. The experiments may contain datasets whereas each dataset contains data of certain type.
The browse experiments page
Experiments are available in a form of a tabular view. The experiment table allows interactive searching for a particular entity using ajax-based engines by searching for experiments using experiment name, description, WP name, etc. The Steps involved in adding a new experiment to the system remain unchanged as described in the Deliverable Report 4.1 however, a slight modification of the interface has been made.
The analysis framework pages:
The Experiment and Dataset Details tool is composed of two parts: Experiment details and Dataset details which functionality will be briefly described in the sections as follow.
Depending on the metadata dataset in the selected experiment, the tool produces a pie chart for the sensory score and temperature data, and bar charts of TVC and bacterial counts of five strains. The sensory score pie chart displays the number of samples for a given sensory score value. The sectors are colored green for the lowest sensory score value and red for the highest value. The colors of the values between the highest and the lowest value are taken from the green to red gradient. This coloring scheme is used for the PCA and HCA tools as well. This part of the tool also presents the TVC and bacterial counts in bar charts.
The Principal Component Analysis tool
The Principal Component Analysis tool is used to perform Principal Component Analysis of the selected dataset (either HPLC, FT-IR or eNose data set). The PCA may help in reducing the number of variables in the dataset and in finding any structures in the relations between variables. In meat spoilage research, the PCA may provide a quick exploratory look at the data.
The main functionality of the PCA tool includes: (i) Generation of the dynamic and interactive PCA plot which can be zoomed in, zoomed out and panned using control buttons (ii) A tool for assigning colors to the sample points according to the values of the selected metadata data set. This functionality can be accessed using the Assign colours button (iii) A tool for including or excluding variables from the PCA calculations. This functionality can be accessed using the Exclude variables button (iv) Two scaling methods: Range-scaling and Auto-scaling which can be accessed using the Scaling method menu (v) The Sample names on/off button used to hide or show sample names on the plot (vi) The Biplot on/off button used to hide or show loadings on the plot (vii) The Reset button used to set the PCA plot to the default view (viii) A popup window with additional metadata information when hovering a mouse pointer over a sample point.
Mathematical modeling and Validation of mathematical/multivariate models to predict meat (beef or pork) quality / spoilage
In the previous section it was evident that a great number of multifactorial experiments were designed and performed during the project period. It was also evident that most instrumental and /or analytical techniques were exchanged among the project partners. This included a Raman Spectrograph supplied by the University of Manchester to the Agricultural University of Athens and a Videometer device supplied by Videometer to AUA, while personnel was exchanged between AUA and the University of Manchester, AUA and Naple, Videometer and AUA, and Cranfield and AUA.
In order to achieve the validation studies, several models were developed based on various analytical techniques applied within WP1 and WP3 such as E-nose, HPLC, FTIR, and videometer. Each of the developed models was validated against a testing dataset and the model performance was assessed based on how accurately the model can classify the freshness of a given sample compared to sensorial evaluation and microbiological enumeration. Models evaluation report was generated which help to identify the most suitable modeling approach for a particular analytical platform, as well as the possibility to improve the model performance by developing prediction models based on a combined output from two or more experimental platforms.
Different models have been developed in this project; two indicative cases are reported below;
In the 1st case (Panagou et al. 2011); A series of partial least squares (PLS) models were employed to correlate spectral data from FTIR analysis with beef fillet spoilage during aerobic storage at different temperatures (0, 5, 10, 15, and 20oC) using the dataset presented by Argyri et al. (2010). The performance of the PLS models was compared with a three- layer feed-forward artificial neural network (ANN) developed using the same dataset. FTIR spectra were collected from the surface of meat samples in parallel with microbiological analyses to enumerate total viable counts. Sensory evaluation was based on a three-point hedonic scale classifying meat samples as fresh, semi-fresh, and spoiled. The purpose of the modeling approach employed in this work was to classify beef samples in the respective quality class as well as to predict their total viable counts directly from FTIR spectra. The results obtained demonstrated that both approaches showed good performance in discriminating meat samples in one of the three predefined sensory classes. The PLS classification models showed performances ranging from 72.0 to 98.2% using the training dataset, and from 63.1 to 94.7% using independent testing dataset. The ANN classification model performed equally well in discriminating meat samples, with correct classification rates from 98.2 to 100% and 63.1 to 73.7% in the train and test sessions, respectively. PLS and ANN approaches were also applied to create models for the prediction of microbial counts. The performance of these was based on graphical plots and statistical indices (bias factor, accuracy factor, root mean square error). Furthermore, results demonstrated reasonably good correlation of total viable counts on meat surface with FTIR spectral data with PLS models presenting better performance indices compared to ANN.
In the 2nd case (Argyri et al. 2010); A machine learning strategy in the form of a multilayer perceptron (MLP) neural network was employed to correlate Fourier transform infrared (FTIR) spectral data with beef spoilage during aerobic storage at chill and abuse temperatures. Fresh beef fillets were packaged under aerobic conditions and left to spoil at 0, 5, 10, 15, and 20C for up to 350h. FTIR spectra were collected directly from the surface of meat samples, whereas total viable counts of bacteria were obtained with standard plating methods. Sensory evaluation was performed during storage and samples were attributed into three quality classes namely fresh, semi-fresh, and spoiled. A neural network was designed to classify beef samples to one of the three quality classes based on the biochemical profile provided by the FTIR spectra, and in parallel to predict the microbial load (as total viable counts) on meat surface. The results obtained demonstrated that the developed neural network was able to classify with high accuracy the beef samples in the corresponding quality class using their FTIR spectra. The network was able to classify correctly 22 out of 24 fresh samples (91.7%), 32 out of 34 spoiled samples (94.1%), and 13 out of 16 semi-fresh samples (81.2%).
In general during this project fresh meat either beef or pork was stored under aerobic and modified atmospheres with or without the addition of volatiles compounds from essential oils; these samples were stored in a wide range of temperatures from 0 to 20o C. During storage, packages were subjected to microbiological analyses for the determination
(i)of Total viable counts, pseudomonads, Enterobacteriaceae, yeasts, Brochothrix thermosphacta and lactic acid bacteria.
(ii)to sensory analysis by a four-member taste panel for the characterisation of meat samples in three pre-defined quality classes, namely Fresh (F), Semi-fresh (SF), and Spoiled (S).
(iii)(date were collected form instruments of vibrational spectroscopy e.g. FTIR, RAMAN, fluorescent spectroscopy, multispectral images and electronic nose (iv) metabolic data were collected from the sample using HPLC and GCMS instruments.
Validation of beef (minced) Quality-using FTIR
The classification table resulting from the FDA provided 100 % correct classification and 76.3% correct classification when cross-validated. Even though 23.7% of samples could not be cross-validated, no fresh (F) sample was reclassified as spoiled (S) or vice-versa. The same approach was followed taken the type of packaging as a factor. PCs resulting from the PCA were subjected to a FDA based on the defined packaging type (air, MAP, and active packaging) constituting the dependent variable. The classification table resulting from the FDA provided 100% correct classification and 92.5% correct classification when cross-validated.
(b) Validation of beef (fillets) Quality-using FTIR
Sensory evaluation was performed during storage and samples were attributed into three quality classes namely fresh (F), semi-fresh (SF), and spoiled (S). A neural network was designed to classify beef samples to one of the three quality classes based on the biochemical profile provided by the FTIR spectra, and in parallel to predict the microbial load (as total viable counts, TVC) on meat surface. The network consisted of an input layer with seven input nodes for storage temperature, sampling time, and the five principal components (PCs). The output layer consisted of two nodes, one for the quality class (Fresh, Semi-fresh, Spoiled), and another for the predicted total viable counts of the beef fillet sample. The classification accuracy of the neural network was determined by the number of correctly classified samples in each sensory class divided by the total number of samples in the class. The validation of the neural network in the prediction of TVC for each meat sample analyzed was determined by the bias (Bf) and accuracy (Af) factors, the mean relative percentage residual (MRPE) and the mean absolute percentage residual (MAPR), and finally by the root mean squared error (RMSE) and the standard error of prediction (SEP).
The performance of the MLP network to predict total viable counts in meat samples in terms of statistical indices is presented in Table 3 (Appendix II). Based on the calculated values of the bias factor (Bf) it can be inferred that the network under-estimated TVC in semi-fresh and spoiled samples (Bf less than 1), whereas for fresh samples over-estimation of microbial population was evident (Bf greater than 1). In addition, the values of the accuracy factor (Af) indicated that the predicted total viable counts were 18.1%, 12.2%, and 8.4% different (either above or below) from the observed values for fresh, semi-fresh, and spoiled meat samples, respectively. The standard error of prediction (SEP) index is a relative typical deviation of the mean prediction values and expresses the expected average error associated with future predictions. The lower the value of this index is, the better the capability of the network to predict microbial counts in new meat samples. The value of the index was less than 10% in spoiled samples indicating good performance of the network for microbial count predictions in this class.
(c) Validation of beef (minced) Quality using FTIR and RAMAN
(d) Validation of beef (minced) Quality using multispectral image analysis
Mathematical model development included initially principal components analysis (PCA) for dimensionality reduction of the data set. Subsequently, for qualitative analysis, principal components (PCs) significantly contributing to the variance of the data set were subjected to factorial discriminant analysis (FDA) to predict the class membership of a sample belonging to a previously-defined qualitative group. Finally, for quantitative analysis, PCs significantly contributing to the variance of the data set were regressed using a partial least squares regression (PLS-R) onto the different microbial groups enumerated during storage data in order to predict the microbial load of meat samples directly from spectral data. The first PCA showed that all the bands were found to be significant and thus all of them were used for further analyses. The FDA provided classifications of the samples with an overall correct classification for the validation sensory scores of 71.15%, whilst the provided estimations from the PLS-R model were better regarding the overall classification (86.54%) and each sensory group separately.
(e) Validation of pork (minced) Quality using multispectral image analysis
The quality of stored minced pork meat was monitored using a rapid multispectral imaging device to quantify the degree of spoilage. Bacterial counts of a total of 155 meat samples stored for up to 580 hours have been measured using conventional laboratory methods. Meat samples were maintained under two different storage conditions - aerobic and modified atmosphere packages as well as under different temperatures.
The multispectral imaging device recorded spectra in the visible and the start of the near-infrared area, and these images can be analysed in conjunction with various machine learning and vision techniques. Features were extracted to evaluate the spoilage process of the meat by predicting Total Viable Counts as well as classifying meat pieces into one of three classes, namely fresh, semi-fresh or spoiled, with ground truth being set by a sensory panel. For the multispectral images, an overall classification performance of 76.1% was achieved. For the microbial counts an overall classification performance of 80.0% was achieved. Thus, considering the fact that the electromagnetic area was sampled in only 18 distinct areas, mainly in the visible region, a classification error of 76.1% is a relatively good performance. A very good discrimination between spoiled and fresh pieces of meat was achieved, while semi-fresh meat caused some misclassification, an issue that can be limited if a significantly larger amount of samples could be introduced for analysis. Concluding, in contrast with the retrospective and laborious microbiological analysis, the rapid non-invasive equipment that has been used, requiring no sample preparation, and in which additional parameters have been included in this work such as time, temperature and atmosphere provided a rather promising tool for assessing pork spoilage.
(f) Validation of pork (minced) Quality using FTIR spectral data
A three-class evaluation scheme was employed in this experiment. The first class (Fresh) corresponded to the absence of off-flavors, equal to the reference sample; the second class (Semi-fresh) corresponded to the presence of slight off-flavors but not spoiled (still acceptable quality); and the third class (Spoiled) corresponded to clearly off-flavors development (unacceptable quality). Semi-fresh was the first indication of meat spoilage (incipient spoilage) in which the sample was marginally accepted. For qualitative analysis, PLS discriminant analysis (PLS-DA) was used to develop models allowing the discrimination of meat samples in the selected sensory classes. Good classification was obtained in all sensory classes demonstrating the effectiveness of the method as a rapid screening technique to monitor minced pork spoilage. Specifically, a number of 6 LVs was finally selected as input variables in PLS-DA model development, presenting the highest correct classification (%) in the training (94.0%) and test (88.1%) datasets.
For the training dataset, the PLS-DA approach provided 100% correct classification for spoiled samples, whereas for fresh and semi-fresh the respective number was 93.3% and 86.7%, respectively, representing 1 misclassification out of 15 fresh samples and 2 misclassifications out of 15 semi-fresh samples. PLS regression (PLS-R) models were built for the counts of TVC, Pseudomonas spp., B. thermosphacta and lactic acid bacteria using FTIR responses as input variables and the microbial counts as output variables. Good relationships were found between the results of FTIR and the microbiological analysis.
(g) Validation of pork (minced) Quality using electronic nose (e-nose).
In this part of the study, e-nose was used to obtain volatile fingerprints of minced pork meat during storage in different conditions (i.e. temperature and packaging systems) in an attempt to monitor spoilage. The eight sensors had different responses to the samples tested and this could be attributed to the intrinsic selectivity of the molecular sensing mechanism and to the mass of the molecules that are bounded at the coated surface of the sensors. The volatile patterns collected from the eight sensors signal of e-nose were subjected to Principal Component Analysis (PCA) in order to reduce the dimensionality of the data set. The total variance (100%) of the data set could be explained by eight principal components (PCs) among which the first five accounted for 99% of total variance observed in the experiment.
(h) Validation of beef (fillets) Quality using electronic nose (e-nose).
The volatile patterns collected from e-nose were initially subjected to Principal Component Analysis (PCA) to reduce multi-collinearity (e.g. sensors with overlapping sensitivities) and allow the information to be displayed in a smaller dimension. Subsequently, the scores of the first five principal components accounting for 99% of total variance observed in the experiment were further used as input in Support Vector Machines (SVM) analysis using linear, polynomial and radial basis function (RBF) kernels, to predict the quality of a meat sample that was pre-characterized as fresh (F), semi-fresh (SF) or spoiled (S) from a taste panel. The approach was similar to this mentioned for validation of pork model. Test data were not employed in any step of model development, but they were used exclusively to determine its performance. The classification accuracy of the SVM model was determined as the number of correctly classified samples in each sensory class divided by the total number of samples in the class. Results showed that SVM with RBF kernels provided good discrimination of beef fillet samples regarding their spoilage status at both packaging conditions.
(i) Validation of beef (fillets) Quality using ensemble-based systems and electronic nose (e-nose).
The bagging approach was followed to develop an ensemble-based system for predicting quality based on the sensory evolution scores given by the panel as described earlier. The dataset was split into training and a testing subset. For each of the classifiers forming the ensemble, the training subset was boostrapped to form a subset of 330 samples, which is then divided further into training and a testing subsets in order to perform the grid search optimization process. A total of 200 classifiers were developed to form the ensemble. The ensemble prediction accuracy was calculated using the testing subset and the final output was computed using various aggregation methods. As shown in Table 15 (Appendix II), the bagging approach has improved the overall prediction accuracy by more than 10% when compared to the single classifier performance when assessed using an unseen testing subset, showing a performance of approximately 83%. All aggregation methods applied for output fusing performed equally well. Naive Bayes aggregation showed the best classification performance, with an overall accuracy of 84.10%.
Potential Impact:
Introduction
Economic Impact on Meat industry - a EU Need
The European Union Beef Market; The Economic Impact of the Meat Products Industry begins with an accounting of the direct employment in the various sectors. Meat production encompasses slaughterhouses, packers, company- owned distribution and supply operations and importers. Wholesaling includes the nationwide network of meat and meat product wholesalers and related warehouse and transportation units. Retailing includes locations where meat is consumed on-premise, such as restaurants. Off-premise retail outlets are supermarkets, butchers, warehouse stores, and similar locations. The data comes from a variety of government and private sources. It is sometimes mistakenly thought that initial spending accounts for all of the impact of an economic activity or a product. For example, at first glance it may appear that consumer expenditures for a product are the total sum of the impact on the local economy. However, one economic activity always leads to a ripple effect whereby other sectors and industries benefit from this initial spending. This inter-industry effect of an economic activity can be assessed using multipliers from regional input-output modeling. The economic activities of events are linked to other industries in the state and national economies. The activities required to produce a can of packaged meat from butchering a hog, to packaging, to shipping generate the direct effects on the economy.
Development of the EU beef industry; in order that the beef industry remains strong in a market-oriented environment it was reported that beef quality specifications, in the whole meat chain, will again become the primary market driver, both from a consumer and producer perspective1. For those EU beef value-chain participants that are able and willing to adapt to the new challenges, the future could offer significant opportunities, but still, the way ahead remains essentially uncharted territory with many cross road decision points. The EU will continue to be a dominant beef consumer and domestic production will need to be at substantial levels to meet this demand.
General Impact of SYMBIOSIS-EU project
The Symbiosis project, has a significant impact on the following key communities within the EU: (i) European consumers of meat products, who will benefit from better quality products, with higher levels of food safety (ii) The European food industry, which will be able to deliver improved products, whose quality and safety can be proven scientifically (iii) The food microbiology community, who will gain new tools and new insights into the mechanisms of food spoilage (iv) The research community in general, who will benefit from new software tools for data integration, and a high profile real world example of their technologies in action.
The project contributed to the general policy of the EC to improve the quality of health and safety delivered to the European citizen regarding the food chain. The safety of food products is currently very much in focus, owing to calamities with microbiological outbreaks, dioxin contamination and other threats to human health (https://webgate.ec.europa.eu/rasffwindow/portal/index.cfm?event=notificationDetailandNOTIF_ REFERENCE=2011.0805 and http://www.foodpoisonjournal.com/foodborne-illness-outbreaks/7-children-in-france-remain-in-the-hospital-due-to-e-coli-infection/).
SPECIFIC IMPACTS
Impact for the industry:
Development of simple, effective and inexpensive evaluation of the meat quality and safety sector (WP 4, and WP5);
Development of active packaging systems for preservation of meat with which safety and quality was assured (WP1);
Development of new vibrational spectroscopies, surface chemistry, fluorescent spectroscopy and electrical nose, for the simple and rapid detection of microbial metabolites indicative of specific muscle quality and safety attributes (WP2 and WP3);
Development of advanced diagnostic methods based on intelligent and statistical schemes to evaluate meat quality (WP2-5, WP7-8);
Development of mathematical models to predict shelf life as well as formation of spoilage compounds (WP5 and WP7).
b- Impact for the society:
The website with current information on the importance of microbial metabolites in meat and other project related findings and information for the dissemination to non-specialist laboratories and Small Medium Enterprises (SMEs) (WP2 and WP8);
Develop and distribute user-friendly application software that facilitate practical use of the methods and mathematical models that have been successfully validated in this project, and thereby provide tools that can improve the competitiveness of the European muscle food sector (WP4 and WP5);
The availability of highly effective sensors have broaden the market opportunities for companies involved in manufacture of electronic systems and components, thus having a positive impact on employment within this sector. In addition new vacancies for qualified engineers to use such automated systems will be available in the food industrial sector (WP5, WP7 and WP9).
c- Impact for the science
Provide basic knowledge tools and resources for the interpretation and management of the generated data (e.g. genomic, proteomic, metabolomic and phenotypic data) relevant to microbial behavior in relation to safety and spoilage of food (WP1, WP2 and WP6);
Provide fundamental knowledge (genomics, metabolomics) of the freshness, spoilage and hazards (e.g. biogenic amines, etc) associated with meat products stored under dynamic conditions, enabling the rational design of procedures for reduction of food losses and to increase quality, consumer satisfaction and public health protection (WP2 , WP6 and WP7).
Impact of Specific Work Packages
Impact of WP1: Monitoring meat microbial quality and safety during production, storage and distribution.
The first task toward an integrated understanding of the Food microbial ecology (the meat ecosystem) was to identify and study the most important components. The impact of this wp was very important. It was shown that storage temperature combined with packaging conditions as well as the application of different treatments/antimicrobials induced a selection of the spoilage microbiota even at strain level.
Impact of WP2: Monitoring of major profiling of metabolites (metabolomics) occurring in meat during production, storage and distribution.
This work package plays a major role in offering a better understanding of the spoilage process within meat products. This involves a series of microbiological assessments of meat quality, in order to identify major metabolic profiling of metabolites throughout the complete procedure of production, storage and distribution.
Impact of WP3: Development and application of an image / sensor-based platform to monitor meat quality.
The conventional microbiological analysis of food products usually focuses on ensuring the arrival of safe product to the customer. On the other hand, from the customer point of view, the product should be satisfactory in terms of appearance, odour and taste. One important task within WPs 2 and 3 is the in-depth study of the chemical changes occurring in meat products under different storage conditions responsible of these sensorial changes. This would have a significant impact on the food industry, as it will provide a more detailed scientific understanding of the effects of different packaging solutions, enabling packaging to be optimised according to the needs of the packed product and the supply chain.
Impact of WP4: System integration
The impact of Integration of the prototype with the Symbiosis Online Research Framework SORF, was indeed priceless since it was doable to combine and upload (incorporate) data (heterogeneous) complete experimental dataset files instead of having to manually add datasets entities to the interface.
Impact of WP5: Mathematical modeling of metabolite formation and sensor responses
This work package, which deals with the mathematical analysis of the data collected in earlier work packages, has a profound impact on the food microbiology sector, both in terms of the discovery of new biological knowledge of interest to industry and academia, and also in terms of demonstrating to the community the benefits of multivariate data analysis in their field. Indeed the traditional conventional analysis of data for microbiologists (and biologists in general) has taken the univariate approach to interpreting their data e.g. looking at a single marker or characteristic at a time. Given the much more powerful analytical approaches now available (and used in this project) multivariate techniques are essential if maximum information is to be extracted from the data.
Impact of WP6: Understanding regulation of metabolomics for meat quality.
A full understanding of metabolic pathways in the spoilage dynamic would be practically impossible without the development of genomic-scale metabolic network. This approach relies on the assembling of various data sources pertaining to all the chemical reactions that take place during the spoilage process. Microbiological and chemical data represent the strongest evidence to the occurrence of a metabolic reaction (through the identification of biomarkers relating to this reaction); this is often referred to as the top-down approach.
Impact of WP7: European validation studies with indices of quality and mathematical models from individual countries.
(i) At industrial level: Promising techniques developed in WP4, and successfully applied through WP1, have tested in-site. The ultimate goal of this task was to quantitatively assess the accuracy of applying these methods in a practical, realistic environment.
(ii) At laboratory level: One of the major challenges facing the practical application of results obtained through meat science research is the lack of agreement between data obtained in different countries/laboratories. In this project the advantage of the availability of broad range of scientists from different laboratories in order to provide a better common understanding of microbial metabolite indices has been applied. This would take place through the cross-validation of results obtained from one laboratory by other beneficiaries' labs.
(iii) At computational level: Knowledge gained from the previous two tasks would serve as a refining loop for the mathematical models in order to develop sufficiently accurate models able to lead new biological experiments, and predict dynamic conditions that were not identifiable through traditional methods.
Impact of WP8: Dissemination including development and distribution of application software.
Making the project outputs we have maximized the impact of the project. Indeed the following data are available among the partners
(i)Acquired data: All analytical data collected during the project will be made available to the community via the data integration system developed in WP4.
(ii)Research findings: Information derived from the acquired data using the mathematical and other data analysis techniques employed will be reported though oral presentations at conferences and in peer reviewed journal articles.
(iii)Software developed: The software underpinning the data integration system will be provided to others wishing to set up a data integration system for their own projects. Similarly, software tools implementing the most effective data analysis techniques identified in WP5 will be made available via the project's web page to allow people to carry out their own analysis without special software or skills.
Results transferred to market and to industry
The main results of this research are expected to improve the competitiveness of the European meat industry. For this reason the project has provided a source of information and education of industrial personnel, on the potential of harmonious coexistence of science and new technology, quality and safety assurance of these highly perishable products. Training has been already performed and involved placement of academic staff within the food industry and the placement of industrial staff in academic institutions. This has enabled both sides (researchers and members of the meat industries) to be trained in the requirements of the meat industry and understand the potential for incorporation of these new devices and tools (e.g. multivariate mathematical modeling approaches) into their business.
List of Websites:
http://www.symbiosis-eu.net
http://www.symbiosis-eu.net/
Contact (coordinator) person; George-John NYCHAS gjn@aua.gr or nychas@hol.gr