Rapid and Quantitative Detection of the Microbial Spoilage of Meat by Fourier Transform Infrared Spectroscopy and Machine Learning

Fourier transform infrared (FT-IR) spectroscopy is a rapid,noninvasive technique with considerable potential for applicationin the food and related industries. We show here that this techniquecan be used directly on the surface of food to produce biochemicallyinterpretable "fingerprints." Spoilage in meat is the resultof decomposition and the formation of metabolites caused bythe growth and enzymatic activity of microorganisms. FT-IR wasexploited to measure biochemical changes within the meat substrate,enhancing and accelerating the detection of microbial spoilage.Chicken breasts were purchased from a national retailer, comminutedfor 10 s, and left to spoil at room temperature for 24 h. Everyhour, FT-IR measurements were taken directly from the meat surfaceusing attenuated total reflectance, and the total viable countswere obtained by classical plating methods. Quantitative interpretationof FT-IR spectra was possible using partial least-squares regressionand allowed accurate estimates of bacterial loads to be calculateddirectly from the meat surface in 60 s. Genetic programmingwas used to derive rules showing that at levels of 10⁷ bacteria·g^-1the main biochemical indicator of spoilage was the onset ofproteolysis. Thus, using FT-IR we were able to acquire a metabolicsnapshot and quantify, noninvasively, the microbial loads offood samples accurately and rapidly in 60 s, directly from thesample surface. We believe this approach will aid in the HazardAnalysis Critical Control Point process for the assessment ofthe microbiological safety of food at the production, processing,manufacturing, packaging, and storage levels.

INTRODUCTION

Top

Introduction

The last decade has seen an exponential increase in the consumerdemand for poultry and poultry products, fueled in part by dietaryhealth considerations. Fears over microbiological food safetyissues, especially the incidence of Salmonella spp. (23, 49)and Campylobacter spp. (11, 39), in conjunction with consumerdemand for a product of consistently high quality, have focusedattention on a particular area of the food production industry,namely, the requirement for a rapid (less than a few minutes)and accurate detection system for microbiologically spoiledor contaminated meat (3).

At present, no such technology exists in the food industry withinthe Hazard Analysis Critical Control Point system for the microbiologicalsafety and quality of meat and poultry products (20, 26, 48).

Muscle foods, such as meat and poultry, are described as spoiledif organoleptic changes make them unacceptable to the consumer.These organoleptic characteristics may include changes in appearance(discoloration), the development of off odors, slime formation,changes in taste, or any other characteristic which makes thefood undesirable for consumption (25, 26). While endogenousenzymatic activity within muscle tissue postmortem can contributeto changes during storage (1, 25, 32, 44), it is generally acceptedthat detectable organoleptic spoilage is a result of decompositionand the formation of metabolites caused by the growth of microorganisms(28, 40). The organoleptic changes which take place will alsovary according to the species of microflora present, the characteristicsof the meat, processing methods, product composition, and theenvironment in which the food is stored (25).

Provided that the atmosphere is moist, a consortium of bacteriais responsible for spoilage of meat stored at between -1 and25°C. It is agreed that spoilage organisms belong primarilyto the genus Pseudomonas, and these have been observed to attachmore rapidly to meat surfaces than other spoilage bacteria (25).The other major members of the spoilage flora of meat storedaerobically include Moraxella spp., Psychrobacter spp., andAcinetobacter spp. While the gram-negative motile and nonmotileaerobic rods and coccobacilli generally dominate the spoilagemicroflora of meat, the initial population may also containvarious levels of gram-positive genera, usually representedby micrococci and then lactic acid bacteria and Bronchothrixthermosphacta (13).

To date, in excess of 40 methods have been proposed to measureand to detect bacterial spoilage in meats (13, 26). These includeenumeration methods based on microscopy, ATP bioluminescence,and the measurement of electrical phenomena (10, 45), as wellas detection methods based on either immunological or nucleicacid-based procedures (42). The major drawback with the rangeof protocols available is that they are time-consuming and labor-intensiveand give retrospective information. However, in a modern food-processingenvironment, monitoring procedures need to give results in realtime so that corrective action can be taken as soon as possible.

The ideal method for the on-line microbiological analysis ofmeat would be rapid, noninvasive, reagentless, and relativelyinexpensive, and these requirements can be met via the applicationof a spectroscopic approach, in combination with any appropriatedata deconvolution strategy based on statistics or machine learning.Such statistical methods include partial least-squares (PLS)regression (36), while a popular and powerful series of machinelearning strategies (37) are based on methods of evolutionarycomputing (4), such as genetic algorithms (GAs) (4, 19, 24)and genetic programming (GP) (6, 33). Fourier transform infrared(FT-IR) spectroscopy involves the observation of vibrationsof molecules that are excited by an infrared beam, and an infraredabsorbance spectrum represents a "fingerprint" which is characteristicof any chemical or biochemical substance (18, 43). This techniqueis also very rapid (taking seconds) and has been shown to bea valuable tool for the characterization of axenically culturedbacteria (22, 34, 38, 47), including single-gene knockout strains(41).

With the prior knowledge that organoleptic spoilage is the resultof decomposition and the formation of metabolites caused bythe growth of microorganisms, this information can be exploitedthrough spectroscopic analysis. Therefore, rather than measuringexclusively the presence of bacteria per se on the meat surface,vibrational spectroscopy can also be used to measure biochemicalchanges within the meat substrate, enhancing and acceleratingthe detection of microbial spoilage. The objective of the presentstudy was to establish this technology through a series of experimentsundertaken on chicken breast muscle at room temperature andanalyzed by horizontal attenuated total reflectance (HATR) FT-IRspectroscopy with PLS analysis. Our research demonstrates theutility of a novel analytical approach based on FT-IR that canenhance and accelerate the detection of microbial spoilage,providing rapid, accurate, and quantitative results. PLS methodsare not always transparent as to which variables are used (2),whereas certain other methods use input variables explicitlyand directly in the construction of the relationship betweenthe spectra observed and the property of interest. GA and GPanalyses fall into this category and are used here to suggestthat proteolysis is the key indicator for the onset of spoilage.

MATERIALS AND METHODS

Top

Materials and Methods

Sample preparation.
Fillets of prepacked fresh chicken breast meat were purchasedfrom national retail outlets on the morning of each experiment.No prepreparation of the meat, such as removal of fat or connectivetissue, washing, or inoculation with bacteria, was undertaken.In order to accelerate the spoilage process, the meat was firstweighed and divided aseptically into 30-g subsamples and comminutedfor 10 s in a Moulinex type 505 180-W coffee mill (MoulinexUK Ltd., Birmingham, United Kingdom). The bowl of the coffeemill was washed and dried with a paper towel between samples.The sample was removed from the coffee mill and placed in theupturned lid of a 90-mm-diameter petri dish and pressed manuallyto a thickness of $~$ 5 mm, using the inverted base of a petri dishas the press. A sterile upturned petri dish base was used tocover the prepared sample, and once 25 samples had been obtained,they were randomized, numbered, and stored on the bench topat ambient temperature (typically 22 ± 1°C).

HATR FT-IR spectroscopy.
FT-IR analysis was undertaken using a ZnSe HATR accessory (SpectroscopyCentral Ltd., Warrington, United Kingdom) on a model IFS28 infraredspectrometer (Bruker Ltd., Coventry, United Kingdom) equippedwith a deuterated triglycine sulfate detector. The ZnSe HATRcrystal was capable of 10 external reflections, with the evanescentfield (5) effecting a depth of 1.01 µm (Spectroscopy CentralLtd.). At 1-h intervals, six replicates were individually excisedwith a scalpel from a petri dish sample and placed in contactwith the ZnSe crystal, and a spectrum for each replicate wascollected. The replicates measured approximately 60 by 10 by5 mm and were inverted so that the aerobic upper surface ofthe comminuted meat was placed in intimate contact with theHATR crystal.

The crystal surface was cleaned with distilled water and a softtissue following collection of each spectrum and washed thoroughlywith acetone, rinsed with distilled water, and dried with asoft tissue at the end of each sampling interval. The IBM-compatiblepersonal computer (PC) used to control the IFS28 spectrometerwas also programmed (using OPUS version 2.1 software, runningunder OS/2 Warp, provided by the manufacturers) to collect spectraover the wavenumber range 4,000 to 600 cm^-1. The reference spectrawere acquired from the cleaned blank crystal prior to the presentationof each sample replicate. All spectra were collected in reflectancemode with a resolution of 16 cm^-1, and to improve the signal-to-noiseratio, 256 scans were coadded and averaged. The collection timefor each sample spectrum was 60 s, and a total of 450 spectrawere collected over the series of three experiments, all ofwhich were undertaken within a 42-day period. At each 1-h samplinginterval, a 1-g subsample of meat was also taken and vortexedfor 60 s in 9 ml of 0.9% physiological saline, and the pH wasrecorded. A dilution series was undertaken, and plates of LabM blood agar base (IDG PLC, Lancashire, United Kingdom) werelawned in triplicate with 50 µl of homogenate and incubatedfor 48 h at 25°C, and the total viable counts (TVC) wererecorded as CFU.

Supervised analysis.
ASCII data were exported from the Opus software used to controlthe FT-IR instrument and imported into Matlab version 5.3 (TheMathWorks, Inc., Natick, Mass.), which runs under MicrosoftWindows NT on an IBM-compatible PC. To minimize problems arisingfrom unavoidable baseline shifts, the spectra were scaled sothat the smallest absorbance was set to 0 and the highest wasset to +1 (47).

When the desired responses (targets) associated with each ofthe inputs (spectra) are known, then the system may be supervised.The goal of supervised learning is to find a model that correctlyassociates the inputs with the targets; this is usually achievedby minimizing the error between the target and the model's response(output) (16, 36). For quantitative interpretation of the FT-IRspectra, the multivariate linear regression method of PLS wasapplied as detailed previously (27) following computations givenelsewhere (36). The input (x) data sets for the supervised-learningmethod contained the full HATR FT-IR spectra (441 absorbancesrepresenting a band of 16 wavenumbers) and known log₁₀(TVC)values (y data) from the first two spoilage experiments, andthese were partitioned into training and cross validation setsaccording to our established modeling practices (31, 46). Duringcalibration of the model, the root mean squared (RMS) errorbetween the true and desired levels for the cross validationdata was calculated, and the lowest RMS error from this wasused to find the optimal calibration which would give the bestgeneral predictive model. Following calibration, the PLS modelwas challenged with the independent test set of data from theentirely separate third and final experiment.

Evolutionary computation.
Although PLS is an excellent method for the quantitative analysisof biological systems (36), the information as to which wavenumbers in the infrared spectrum are important is not readilyavailable. The use of PLS, therefore, is often perceived asa "black box" approach to modeling spectra and so has limiteduse for the deconvolution of hyperspectral data in chemicalor biochemical terms. Therefore, in this study, evolutionarycomputation methods based on GAs and GP were employed to aidin the deconvolution of these hyperspectra.

GA.
A GA is an optimization method based on the principles of Darwinianselection (24), where, over a series of generations, a populationof parameter sets evolves until an optimal, or near-optimal,solution to a given problem is found.

First, a population of n objects (chromosomes) is created, eachchromosome containing a string of numbers or binary digits representingthe parameters of the problem to be optimized. The populationis randomized so that n sets of "unique" parameter values canbe evaluated and assigned a fitness value (usually a singlenumerical value). Once all n fitness values have been assigned,the next generation of chromosomes is created. In order forthis new generation to be fitter than the last, principles analogousto those of sexual and asexual reproduction are applied. Usinga stochastic selection method, based on parent fitness, twochromosomes are chosen to reproduce, swapping sections of theirrespective sequences (the probability of a particular parentchromosome being selected for sexual reproduction is proportionalto its fitness, so chromosomes with a respectively high fitnessvalue will have a greater chance of selection). This processcreates two new child chromosomes inheriting characteristicsof their parents. The child chromosomes are then subjected tomutation, where the value of each parameter may be randomlychanged. The probability of this change is normally very small.The process of selection followed by reproduction followed bymutation is then repeated until n new chromosomes are created(i.e., a new population to replace the old). The fitness valueis then evaluated for each of the new chromosomes, and the wholeprocess repeats itself. The algorithm continues until a stoppingcriterion is reached. For example, the criterion may be thata given optimal fitness value is met, a certain number of generationshas passed, or the chromosomes have converged to similar parametervalues.

The GA-multiple linear regression (MLR) wavelength selectionmethodology (9) uses a GA to determine the subset of n wavenumbers,taken from the total spectral data set, which, when appliedto an MLR model, will optimally discriminate between the HATRspectra and the known log₁₀(TVC). All calculations were performedwith in-house software which runs under Microsoft Windows NTon an IBM-compatible PC, and full details of GA-MLR are givenelsewhere (9). Briefly, optimization is achieved by monitoringthe RMS error of prediction for each model. The GA uses two-pointcrossover with mutation (19), operating on a population of binary-encodedchromosomes, each chromosome representing n candidate wavelengths(9); in this study, n was 2, 3, and 5. The selection of parentchromosomes for the next generation is carried out using a rank-basedscheme (D. Whitley, presented at the proceedings of the ThirdInternational Conference on Genetic Algorithms, San Mateo, Calif.,1989). No two identical candidates were allowed in a given population.

GP.
A GP is an application of the GA approach to derive mathematicalequations, logical rules, or program functions automatically(17, 21, 33). Rather than representing the solution to the problemas a string of parameters, as in a conventional GA, a GP usually(as here) uses a tree structure. The leaves of the tree, orterminals, represent input variables or numerical constants.Their values are passed to nodes, at the junctions of branchesin the tree, which perform some numerical or program operationbefore passing on the result further towards the root of thetree. Mutations are performed by selecting a parent and modifyingthe value or variable returned by a terminal or changing theoperation performed by a node. Crossovers are performed by selectingtwo parents and swapping subtrees at randomly selected nodeswithin their trees. The new individuals so generated replaceless fit members of the population chosen probabilisticallyon the basis of their unfitness.

The GP employed the genomic computing software Gmax-bio (AberGenomic Computing, Aberystwyth, United Kingdom), which runsunder Microsoft Windows NT on an IBM-compatible PC. An introductionto Gmax-bio is given elsewhere (29), and the default parametersettings for population size (1,000), mutation, and recombinationrates were used throughout. The operators that were used were+, -, /, *, log₁₀(x), 10^x, >, <, and Tanh(x). The fitnesscalculation used is F = 1/(0.01 + S/B), where the values ofS and B are determined by the FITNESS setting. In this expression,S is a statistic derived from the model which ranges between0 and infinity and B is a normalizing quantity. The value ofB is chosen such that a perfect model yields an F of 100 anda model which performs no better than random chance yields anF of 1.

RESULTS AND DISCUSSION

Top

Results and Discussion

The comminution of samples in order to accelerate the spoilageprocess was successful, as the final mean log₁₀(TVC) of 9.02(Table 1) was an order of magnitude above the 10⁸ CFU g^-1 generallyaccepted as the point at which organoleptic spoilage becomesreadily detectable (13, 15). Using 10⁸ CFU g^-1 as the indicatorfor postspoilage, the average spoilage time over the seriesof experiments was 13.6 h (Table 1). The spoilage of the sampleswithin 24 h at room temperature was anticipated, as comminutionruptures cell walls, releasing a source of nutrients; increasesthe surface-area-to-volume ratio; and distributes bacteria thatwould normally be restricted to the surface throughout the meatsubstrate. The initial mean pH range of fresh samples duringthe three experiments (5.7 to 5.9) was within those describedpreviously in the literature (13, 15). The use of pH as an indicatorof spoilage or remaining shelf life in meats would be insufficient,as the pH fluctuates prior to spoilage, only rising significantlywhen levels of bacteria reach $~$ 10⁸ CFU g^-1. At this level, sensoryspoilage is readily detectable, and this is partly a consequenceof the increase in pH and the production of malodorous substances,such as ammonia, dimethylsulfide, and diacetyl, by the catabolicaction of the resident microflora (13, 25).

View this table:
[in this window]
[in a new window]
TABLE 1. Data matrix of results from three spoilage experiments^a

The 150 FT-IR spectra from experiment 1 are shown in Fig. 1and illustrate the reproducibility of both HATR as a samplingmethod and the experimental protocol that was undertaken overa period of 6 weeks. Typical FT-IR spectral data from the 1,750-to 700-cm^-1 wave number range from measurements of meat at thepre- and postspoilage stages are shown in Fig. 2. These spectraare from chicken breast meat carrying $~$ 7 · 10⁶ and $~$ 2 ·10⁹ CFU g^-1, respectively, and are both data rich and not biasedto any particular group of chemicals associated with a particulargroup of metabolites. Furthermore, the spectra are complex andmultidimensional in nature, so they do not easily lend themselvesto simple visual interpretation; this is compounded by the factthat the data set for all three experiments is substantial,with a total of 450 spectra, each containing 441 wave numbers.However, with the advent of modern machine learning approaches,the opportunity now exists to analyze such complex high-dimensionalspectral patterns (7, 46) and to extract an answer to a questionof biological interest with much lower dimensionality, i.e.,"What is the bacterial load on the meat surface?"

View larger version (25K):
[in this window]
[in a new window]
FIG. 1. All 150 HATR absorbance spectra from both fresh and spoiled meats from experiment 1. These illustrate the reproducibility of both HATR FT-IR spectroscopy and the preparation method employed throughout the series of experiments. The amide I and amide II vibrations from proteins and the CH_x vibrations from fatty acids are indicated. The box indicates where the most variance in these spectra occur and hence where spoilage signals are likely to be seen.

View larger version (29K):
[in this window]
[in a new window]
FIG. 2. Typical FT-IR absorbance spectra from pre- and postspoilage chicken. Also shown is the Pearson correlation coefficient (R) between the FT-IR absorbances (in experiments 1 and 2) and the log₁₀(TVC). The asterisks indicate peaks that are attributable to amide I (1,640 cm^-1), amide II (1,550 cm^-1), and amine (1,240 and 1,088 cm^-1) vibrations.

Therefore, as described above, the supervised-learning methodof PLS regression was calibrated and cross validated with theFT-IR spectral data and the known log₁₀(TVC) values from experiments1 and 2 (Table 2 shows the details and TVC levels) before beingchallenged by the independent and "unseen" test set of datafrom experiment 3. The plots of the estimates versus the knownlog₁₀(TVC) (Fig. 3) show that the FT-IR and PLS predictionswere virtually indistinguishable from the expected log proportionalfit [i.e., log(y) = log(x)] and so show that this approach canbe used to accurately assess the spoilage status of meat. Ascan be seen in Fig. 3 and Table 2, the lowest level of spoilageencountered was $~$ 2 · 10⁶ CFU g^-1, and this necessarilyrestricts the detection limit. The TVC for chicken immediatelypostslaughter is 10³ cm^-2, rising to 10⁴ to 10⁵ cm^-2 after packaging(26). That PLS gave accurate results at 2 · 10⁶ CFU g^-1suggests that it will be possible to reach lower levels, andthis will be the subject of further study with freshly killedchickens. From Fig. 3, it is evident that the spectra obtainedby direct FT-IR analysis of meat do contain biochemical informationthat allows correlation with the spoilage status of the chicken,for data used to produce the PLS model and, more importantly,for data from a completely new experiment. The obvious questionthat needs to be addressed is that of which biochemical speciesthe FT-IR is measuring that are related to the spoilage statusof the chicken.

View this table:
[in this window]
[in a new window]
TABLE 2. Log₁₀(TVC) of bacteria acquired from comminuted meat samples from three experiments^a

View larger version (33K):
[in this window]
[in a new window]
FIG. 3. Estimates from PLS versus the true log₁₀(TVC). The data points are the averages of the six measurements, and the error bars show standard deviations. The RMS errors in these measurements are 0.15, 0.23, and 0.27 log units for the calibration, cross validation, and independent test sets, respectively.

The Pearson correlation coefficients between the absorbancesat each wave number in the FT-IR spectra from experiments 1and 2 and the log₁₀(TVC) were calculated and are also plottedin Fig. 2. It can be seen that most peaks from 1,500 to 700cm^-1 are positively correlated with spoilage, but no singlepeak appears uniquely dominant; this necessarily means thatit is difficult to pinpoint the cause of microbial spoilageto a single (or a small group of) biochemical species usingthis correlation approach. Therefore, GAs and GPs were evolvedto discriminate qualitatively between meat carrying <10⁷and $>=$ 10⁷ bacteria (as TVC) per cm².

GA-MLR was applied so as to extract subsets of two, three, andfive wave numbers that could discriminate between fresh (<10⁷bacteria/cm²) and spoiled ( $>=$ 10⁷ bacteria/cm²) chicken. Becausethe starting population for each GA run was random, 60 GA-MLRruns were performed, and the following subsets were found tobe optimal for selecting just two or three wave numbers, respectively:(1,096, 1,227 cm^-1) and (1,312, 1,235, 1,088 cm^-1). When thealgorithm was used to look for five wave numbers, it was foundthat the degree of discrimination did not improve compared withselecting subsets of three, and no consistent areas of the FT-IRspectra were found to be dominant in the GA expressions; however,vibrations at 1,096 and 1,305 cm^-1 were found within the bestsubsets.

GP analyses (i) using the same 10⁷-bacteria/cm^-1 threshold asabove and (ii) evolved to predict the log₁₀(TVC) levels producedtrees which could easily discriminate between fresh and spoiledchicken and quantify the level of spoilage, respectively; atypical GP parse tree is shown in Fig. 4. As with the GAs, theinitial populations were produced randomly; therefore, 10 separateGPs were evolved. For the threshold GP analysis, the numberof times each input (wave number) was used for the 10 evolvedpopulations was calculated and plotted against the wave numberof the infrared light (Fig. 5). Figure 5 clearly shows thatthe dominant area of the spectra for discriminating betweenfresh (<10⁷ bacteria/cm²) and spoiled ( $>=$ 10⁷ bacteria/cm²)chicken was 1,088 to 1,096 cm^-1; moreover, these wave numberswere also selected by the GA-MLR method. The functional groupvibration in the region 1,088 to 1,096 cm^-1 is ascribable toC-N stretching, most plausibly from amines (14, 35).

View larger version (13K):
[in this window]
[in a new window]
FIG. 4. Typical GP tree evolved to discriminate between chicken carrying <10⁷ and $>=$ 10⁷ bacterial counts. The use of the logistic function Pr{Y} = 1/(1 + exp^{(-Model Expression)}), defines a maximum-likelihood (Pr{Y}) decision boundary for an output being either false (0) or true (1).

View larger version (28K):
[in this window]
[in a new window]
FIG. 5. Frequency plot of the number of times an input was used in 10 independent GPs, evolved to discriminate between chicken carrying <10⁷ and $>=$ 10⁷ bacterial counts (encoded as 0 and 1, respectively).

The most intense peaks that appear in fresh meat are the amideI (C $-$ O vibration at 1,640 cm^-1) and amide II (NH deformationat 1,550 cm^-1) bands from proteins and peptides, and from thePearson correlation coefficients, the amide II band is the onlyvibration that is negatively correlated with spoilage (Fig.2). This strongly suggests that the protein content of the meatwas decreasing during spoilage. By contrast, the peaks at 1,240and 1,088 cm^-1, which are both ascribable to C-N stretchingfrom amines from free amino acids, are positively correlated.Indeed, the rule in Fig. 4 shows that spoilage can be ascribedsimply to the ratio of 1,096 to 1,683 cm^-1 from these vibrationsfrom amines and amides, respectively.

Plots of the absorbances of these vibrations versus the timefor the second experiment are shown in Fig. 6. It is clear thatthe amide I and II bands are constant, although the amide IIband does decrease very slightly after 16 h while the peaksat 1,240 and 1,088 cm^-1 start to increase significantly after16 h. It is noteworthy that the onset of spoilage, as characterizedby a TVC of >10⁸ g^-1, was at 17 h, and this was the pointat which the absorbance due to free amines started to increase.This was also found to be the case for experiments 1 and 3 (datanot shown). These correlations, and the fact that the GAs andGPs both pick the region 1,088 to 1,096 cm^-1 as the most significantarea of the FT-IR spectra for the prediction of spoilage ofchicken which is attributable to free amino acids, makes itclear that the most significant metabolic process that occursat spoilage is the start of proteolysis. This is indeed highlylikely, since it is known that spoilage in meat is most frequentlyassociated with the postglucose utilization of amino acids byaerobic microorganisms, such as pseudomonads, and the onsetof the enzymatic degradation of proteins and peptides, leadingto the production of free amino acids (8, 12, 40).

View larger version (23K):
[in this window]
[in a new window]
FIG. 6. Plots of selected IR vibrations versus time for experiment 2. The selected vibrations are amide I (C $-$ O vibration at 1,640 cm^-1) and amide II (NH deformation at 1,550 cm^-1) bands and C-N stretching from amines at 1,240 and 1,088 cm^-1. Note the increase in the absorbance of 1,088 cm^-1 at 17 h, which corresponds to the point at which the onset of spoilage occurs (Table 1).

In conclusion, FT-IR spectroscopy, in combination with appropriatemachine learning methods, presents itself as a novel methodfor the quantitative detection of food spoilage. Using FT-IR,we were able to acquire a metabolic snapshot (30) and quantify,noninvasively, the microbial loads of food samples accuratelyand rapidly (within 60 s) directly from the sample surface.We believe that this approach has considerable potential forfurther development and will aid both the food safety regulatorybodies and the Hazard Analysis Critical Control Point system.In particular, we will conduct future studies testing our methodfor quantifying the numbers of spoilage organisms on musclefoods at the production, processing, packaging, and storagelevels.

ACKNOWLEDGMENTS

We are indebted to the Agri-Food and Engineering and BiologicalSystems Committees of the United Kingdom BBSRC for supportingthis work.

FOOTNOTES

* Corresponding author. Mailing address: Institute of Biological Sciences, University of Wales, Aberystwyth, Ceredigion SY23 3DD, United Kingdom. Phone: 44 (0)1970 621947. Fax: 44 (0)1970 621947. E-mail: rrg{at}aber.ac.uk.

REFERENCES

Top