Rapid Identification of Closely Related Muscle Foods by Vibrational Spectroscopy and Machine Learning

Abstract

Muscle foods are an integral part of the human diet and during the last few decades consumption of poultry products in particular has increased significantly. It is important for consumers, retailers and food regulatory bodies that these products are of a consistently high quality, authentic, and have not been subjected to adulteration by any lower-grade material either by accident or for economic gain. A variety of methods have been developed for the identification and authentication of muscle foods. However, none of these are rapid or non-invasive, all are time-consuming and difficulties have been encountered in discriminating between the commercially important avian species. Whilst previous attempts have been made to discriminate between muscle foods using infrared spectroscopy, these have had limited success, in particular regarding the closely related poultry species, chicken and turkey. Moreover, this study includes novel data since no attempts have been made to discriminate between both the species and the distinct muscle groups within these species, and this is the first application of Raman spectroscopy to the study of muscle foods. Samples of pre-packed meat and poultry were acquired and FT-IR and Raman measurements taken directly from the meat surface. Qualitative interpretation of FT-IR and Raman spectra at the species and muscle group levels were possible using discriminant function analysis. Genetic algorithms were used to elucidate meaningful interpretation of FT-IR results in (bio)chemical terms and we show that specific wavenumbers, and therefore chemical species, were discriminatory for each type (species and muscle) of poultry sample. We believe that this approach would aid food regulatory bodies in the rapid identification of meat and poultry products and shows particular potential for rapid assessment of food adulteration.

Full text

Rapid identification of closely related muscle foods by vibrational
spectroscopy and machine learning
David I. Ellis,
ab
David Broadhurst,
ab
Sarah J. Clarke
b
and Royston Goodacre*
a
Received 11th August 2005, Accepted 11th October 2005
First published as an Advance Article on the web 26th October 2005
DOI: 10.1039/b511484e
Muscle foods are an integral part of the human diet and during the last few decades consumption
of poultry products in particular has increased significantly. It is important for consumers,
retailers and food regulatory bodies that these products are of a consistently high quality,
authentic, and have not been subjected to adulteration by any lower-grade material either by
accident or for economic gain. A variety of methods have been developed for the identification
and authentication of muscle foods. However, none of these are rapid or non-invasive, all are
time-consuming and difficulties have been encountered in discriminating between the
commercially important avian species. Whilst previous attempts have been made to discriminate
between muscle foods using infrared spectroscopy, these have had limited success, in particular
regarding the closely related poultry species, chicken and turkey. Moreover, this study includes
novel data since no attempts have been made to discriminate between both the species and the
distinct muscle groups within these species, and this is the first application of Raman spectroscopy
to the study of muscle foods. Samples of pre-packed meat and poultry were acquired and FT-IR
and Raman measurements taken directly from the meat surface. Qualitative interpretation of
FT-IR and Raman spectra at the species and muscle group levels were possible using discriminant
function analysis. Genetic algorithms were used to elucidate meaningful interpretation of FT-IR
results in (bio)chemical terms and we show that specific wavenumbers, and therefore chemical
species, were discriminatory for each type (species and muscle) of poultry sample. We believe that
this approach would aid food regulatory bodies in the rapid identification of meat and poultry
products and shows particular potential for rapid assessment of food adulteration.
Introduction
Whilst muscle foods, which include meat and poultry, play an
important part in the human diet and have done so for several
thousand years,
1
significant numbers of consumers eschew
this group of foods, either in part or as a whole, for religious,
moral, cultural or dietary health considerations. The dietary
health considerations for example could involve factors as
simplistic as variations in fat content between different species
or by perceived health risks surrounding large-scale food
safety issues such as BSE and foot-and-mouth disease.
2,3
As
well as these considerations, it is also important for consumers,
retailers and food regulatory bodies that the products
concerned are of a high quality, authentic, and have not been
subjected to adulteration by any lower-grade material either by
accident or for economic gain.
4–9
Therefore, the ability to
determine the authenticity of any foodstuff is important
10–14
and in muscle foods, species identification has been recognised
as a significant issue.
15–18
At present, a variety of methods have been developed for
the identification and authentication of meat and poultry
products. These include methods based on the detection of
DNA or RNA,
19,20
immunological,
21–23
electrophoretic
24–26
and chromatographic
27
techniques. The disadvantage with
these current techniques however, is that they are time con-
suming, labour intensive and require a considerable amount of
background training. Conversely, vibrational spectroscopic
techniques (including mid-infrared, near-infrared and Raman)
are rapid, reagentless, non-destructive and would be ideally
suited to this type of analysis.
28–33
Several studies have applied
infrared (both mid and near) spectroscopic techniques and
visible spectroscopy to the problem of species identification
and authenticity in meat and poultry.
17,18,34
These have
resulted in varying degrees of success, and none of these
methods have been successful in differentiating between the
closely related species chicken and turkey. In addition, no
study has investigated Raman spectroscopy as a method for the
speciation of muscle foods. Indeed, no vibrational spectro-
scopic method has been shown to discriminate between distinct
muscle groups within a species.
The consumption of poultry has risen dramatically in recent
years and this is partly as a consequence of the popularity of
convenience foods, which have significantly increased its
commercial value.
35
The objective of the series of experiments
undertaken for this study was to apply vibrational spectro-
scopic methods, namely HATR (horizontal attenuated total
reflectance) FT-IR and Raman spectroscopy, to the problem
of authenticity in comminuted muscle foods; the samples
a
School of Chemistry, University of Manchester, PO Box 88, Sackville
Street, Manchester, UK M60 1QD.
E-mail: Roy.Goodacre@manchester.ac.uk; Fax: +44 (0) 161 306 4519;
Tel: +44 (0) 161 306 4480
b
Institute of Biological Sciences, University of Wales, Aberystwyth,
Ceredigion, UK SY23 3DD
PAPER www.rsc.org/analyst | The Analyst
1648 |
Analyst
, 2005, 130, 1648–1654 This journal is ßThe Royal Society of Chemistry 2005

(in particular those from poultry) were minced so as to make
them visibly indistinguishable from each other.
Materials and methods
Sample preparation
Samples of pre-packed meat and poultry were purchased from
national retail outlets on the morning of each experiment. For
the preliminary analyses these included chicken (skinless breast
fillets), pork (boneless steaks), turkey (skinless breast fillets),
lamb (neck fillets), beef (rump steaks) and for the subsequent
experiments, chicken (skinless breast fillets and legs with skin)
and turkey (skinless breast fillets and legs with skin). These
were stored at 4 uC until required. For reasons of brevity
samples for the preliminary FT-IR analysis were not homo-
genised but cut into solid samples measuring approximately
60 610 610 mm.
For subsequent experiments, in order to ‘disguise’ the
poultry samples (that is to say make them visibly identical),
individual samples were first weighed aseptically into 30 g
sub-samples (after removal of skin if this was present) and
comminuted for 10 s in a Moulinex type 505, 180 W coffee mill
(Moulinex UK Ltd, Birmingham, UK). The bowl of the coffee
mill was washed and dried with a paper towel between each
sample. The sample was removed from the coffee mill and
placed into the upturned lid of a 90 mm Petri dish and pressed
manually to a thickness of y5 mm using the inverted base of a
Petri dish as the press. A sterile upturned Petri dish base was
used to cover each prepared sample. A total of 4 Petri dish
samples were prepared for each group (e.g. chicken breast).
These were then analysed immediately following preparation.
HATR FT-IR spectroscopy
FT-IR analysis was undertaken using a ZnSe HATR accessory
(Spectroscopy Central Ltd, Warrington, UK) on a Bruker
IFS28 infrared spectrometer equipped with a DTGS (deuter-
ated triglycine sulfate) detector (Bruker Ltd, Coventry, UK) as
described elsewhere.
30
For preliminary analysis, 12 whole-meat
replicates were individually analysed from five species; beef,
lamb, pork, chicken and turkey. Each whole-meat replicate
was placed in intimate contact with the HATR crystal, with no
preference given to which surface of the sample was to be
measured, and a FT-IR absorbance spectrum collected. For
the analysis of poultry species in subsequent experiments,
15 replicates of each type (chicken breast, chicken leg, turkey
breast and turkey leg) were individually excised by scalpel from
the prepared Petri dish samples and the upper surface placed
in intimate contact with the HATR crystal and a spectrum
collected.
The crystal surface was cleaned with distilled water and a
soft tissue following collection of each spectrum and washed
thoroughly with acetone, rinsed with distilled water and dried
with a soft tissue at the end of each sampling interval. The
IBM-compatible PC used to control the IFS28 spectrometer
was also programmed (using OPUS version 2.1 software
running under OS/2 Warp provided by the manufacturers)
to collect spectra over the wavenumber range 4000 cm
21
to
600 cm
21
. The reference spectra were acquired from the
cleaned blank crystal prior to the presentation of each sample
replicate. All spectra were collected in absorbance mode with a
resolution of 16 cm
21
, and to improve the signal-to-noise ratio
256 scans were co-added and averaged. Collection time for
each sample spectrum was 60 s and a total of 60 spectra
were collected for the preliminary analysis and 120 for the
comminuted poultry experiments. Each experiment was
undertaken in duplicate. Any CO
2
peaks were removed prior
to analysis. Typical FT-IR spectra from poultry are shown in
Fig. 1; each spectrum is represented by 441 wavenumbers.
Raman spectroscopy
Raman spectra were collected using a near infrared diode laser
with an excitation at 785 nm, using a Renishaw 2000 Raman
probe system together with the Renishaw WiRE Grams
software package and a CCD detector (Renishaw PLC,
Gloucestershire, UK)
36
as described elsewhere.
31,37,38
The
probe had an optimal focusing distance of 12 mm from the
sampling point and the laser power was set at 78 mW to
measure spectra of chicken and turkey samples. Pure ethanol
was used as a standard (it has a characteristic Raman shift
from the C–C–O vibration at 882 cm
21
) and reference spectra
were collected at the start of each experiment. Fifteen spectra
(measurements) were randomly taken from each of the four
samples, namely chicken breast and leg and turkey breast and
leg, which were repeated in triplicate over three days. Raman
spectra were collected for 10 seconds and 1 accumulation over
the wavenumber range 100 cm
21
to 3000 cm
21
. Typical
Raman spectra from poultry are shown in Fig. 2.
Cluster analysis
For FT-IR, ASCII data were exported from the Opus software
used to control the FT-IR instrument and imported into
Matlab version 6.1 (The MathWorks, Inc., Natick, MA) which
runs under Microsoft Windows NT on an IBM-compatible
PC. To minimise problems arising from unavoidable baseline
shifts the spectra were scaled so that the smallest absorbance
was set to 0 and the highest to +1.
39
In the case of Raman
measurements, spectra were converted to multifiles, cosmic
rays removed, and to account for photon count differences the
spectra were scaled such that the offset = 0 and the height of
the first line (where the laser line is cut out by the holographic
filter) at 250 cm
21
=1.
Matlab was then used to perform principal components-
discriminant function analysis (PC-DFA) on both Raman and
FT-IR spectral data sets as described elsewhere.
39
Briefly, the
initial stage of the cluster analyses involved the reduction of
the multidimensional spectroscopic data by principal compo-
nents analysis (PCA).
40
PCA is a well known technique for
reducing the dimensionality of multivariate data whilst
preserving most of the variance, and Matlab was employed
to perform PCA according to the NIPALS algorithm.
41
Discriminant function analysis (DFA; also known as canonical
variates analysis (CVA)) then discriminated between groups on
the basis of the retained principal components (PCs) and some
a priori class structure.
42
Two strategies were used:
(a) For the analysis of chicken, pork, turkey, lamb and beef
using FT-IR spectroscopy the class structure was two groups
This journal is ßThe Royal Society of Chemistry 2005
Analyst
, 2005, 130, 1648–1654 |1649

per meat type relating to the two duplicate experiments. Thus
there were 10 classes in total.
(b) For the analysis of chicken breast, chicken leg, turkey
breast and turkey leg using FT-IR and Raman spectroscopies
the class structure was one per muscle and species type. That is
to say there were 4 classes in total. In order to validate this
process, data from a separate experiment were projected into
PC-DFA space constructed from training data from other
independent experiments.
Genetic algorithms
Having shown that there was discriminatory information
within the FT-IR fingerprints, computational heuristic search
methods, such as genetic algorithms (GA) were used to aid the
discovery of important biochemical features in these spectra.
In this study a GA was applied to determine the subset of n
wavenumbers, taken from the FT-IR data matrix, which when
applied to a discriminant multiple linear regression (D-MLR)
model
43
would optimally distinguish between two selected data
Fig. 1 Typical FT-IR absorbance spectra of chicken and turkey for leg and breast muscles. Also shown are the wavenumbers selected by GA as
the most discriminatory bands. See Table 1 and text for details.
Fig. 2 Typical Raman spectra of chicken and turkey for leg and
breast muscles.
Table 1 Confusion matrices showing the predictions from PC-DFA
projection analysis of the muscle and species type from Raman spectra
Identities
Predictions
Chicken
breast
Chicken
leg
Turkey
breast
Turkey
leg
Training data Chicken breast 30 0 0 0
Chicken leg 0 29 0 1
Turkey breast 0 0 30 0
Turkey leg 0 2 0 28
Test data Chicken breast 11 1 3 0
Chicken leg 1 13 0 1
Turkey breast 2 0 13 0
Turkey leg 0 2 0 13
1650 |
Analyst
, 2005, 130, 1648–1654 This journal is ßThe Royal Society of Chemistry 2005

groups. Initially GA D-MLR analysis was performed in order
to discriminate between the two general muscle types and the
two species. The analysis was then extended such that four GA
models were used to discriminate between each of the four
muscle groups in turn (chicken breast, chicken leg, turkey
breast and turkey leg) and the remaining 3 types. Training data
were from the first FT-IR experiment, whilst test data were
from the second independent experiment. All calculations were
performed using in-house software written in C++ running
under Microsoft Windows NT on an IBM-compatible PC and
optimization was achieved by monitoring the residual mean
square error of prediction (RMSEP) for each model. Full
details of the GA-MLR algorithm are given in ref. 43, and a
previous application using this technique to discriminate
between control and salt-treated tomato fruit is described in
Johnson et al.
44
In this study the GA used proportional selection, and two-
point crossover recombination with mutation, operating on a
population of binary-encoded chromosomes, each chromo-
some representing vcandidate wavelengths. The parameter v
can be set to any integer value, between 2 and the total number
of wavelengths used, prior to the execution of any single GA
run. In order to select optimally the value of v, a set of
GA experiments are performed where vis varied between 2
and v
max
(v
max
being the minimum number of wavenumbers
allowed before overtraining in the regression model occurs).
The probability of mutating a given chromosome after
recombination was set to 0.2, and the probability of changing a
bit from a 0 to 1 (or vice versa), once a chromosome is selected
for mutation, was set to 0.01. No two identical candidates
were allowed in a given population and the top ten percent of
each generation are automatically transferred unchanged to
the next generation. A total of 100 independent GA runs were
performed for each discrimination model in order to statisti-
cally validate the results of each genetic search.
Results and discussion
Preliminary analysis
Results from the preliminary FT-IR analysis on five different
species, where samples were presented intact, illustrate that all
5 species could be clearly differentiated using PC-DFA (Fig. 3).
Furthermore, the discrimination between mammalian and
avian muscle types is evident as is the observation that chicken
and turkey are closely related to each other, as are pork and
lamb, whilst beef appears to be the least related to any of the
other species. It is known that there are distinct differences
between avian breast and mammalian muscle types, related, in
the formers case, to adaptations to flight. Mammalian muscle
for example has higher levels of myoglobin, numbers of
mitochondria and a higher capillary density than those
observed in avian breast muscle, whilst the avian muscle is
adapted to obtain energy from large glycogen stores.
45
Whilst
the ability to discriminate between these 5 species using FT-IR
spectroscopy does not appear to be problematic in the present
study, previous workers have failed to discriminate unequi-
vocally between the closely related avian species.
17,18,34
The
next stage was therefore to investigate this further with
different muscle types using FT-IR and Raman spectroscopies.
Analysis of poultry using Raman spectroscopy
In order to disguise the meat species, chicken and turkey meat
from breast and leg muscles were comminuted as described
above. All samples were analysed pure and no attempt in the
present study was made to quantify mixtures of these muscle
and species types; in part because of the conclusions made
by ref. 17, 18 and 34 who found that whilst quantification of
mixtures of different species (viz. pork from beef) was possible,
the discrimination between poultry species was intractable.
All four meats were analysed first by Raman spectroscopy
as described above. The Raman spectra are shown in Fig. 2,
and visually are more complex than the FT-IR spectra (Fig. 1).
It is clear that these Raman spectra contain a strong baseline
shift that is likely to have arisen from fluorescence which
has been produced when these samples have been analysed.
Indeed, this is a common occurrence when analysing biological
Fig. 3 (A) Typical FT-IR spectra from five meat species studied
collected without comminution. (B) PC-DFA plot on FT-IR spectra
showing the relationship between beef (B), lamb (L), pork (P), chicken
(C) and turkey (T). Data were collected in duplicate with six samples
from each species per experiment. For PC-DFA a total of 20 PCs were
used with the a priori knowledge of 2 classes per species.
This journal is ßThe Royal Society of Chemistry 2005
Analyst
, 2005, 130, 1648–1654 |1651

material even with a near infrared laser.
36,47
Whilst the Raman
spectra are qualitatively similar, subtle quantitative differences
can be seen. However, the visual inspection of all these high
dimensional spectra is not possible and hence cluster analysis
was used.
As detailed above, each meat was analysed 15 times, 3 times
over a period of 3 d. Data from two of those days were used in
the construction of PC-DFA using the a priori information
about the muscle type and poultry species (i.e., there were four
groups each containing 30 Raman spectra). After PC-DFA
was performed, the test set from the remaining day was
projected first into the PCA space and then the projected PCs
projected into DFA space as detailed elsewhere.
46,47
The
resulting PC-DFA projection plot is shown in Fig. 4, where it
can be clearly seen that in the first discriminant function the
major discrimination was between leg and breast muscle
irrespective of species type. This is perhaps expected given the
biochemical nature of these two muscle types. On closer
inspection of Fig. 4 it is also possible to see discrimination
between the poultry species in the second discriminant
function; note that the scale in DF2 is much smaller than that
for DF1 indicating that this is a much smaller spectral
difference. For ease of interpretation, the PC-DFA has been
converted into confusion matrices for both the training and
test data (Table 1). It can be seen for the test data that only 2
of the 60 spectra are recovered to the wrong muscle type, while
8/60 are recovered to the wrong species type. All other spectra
were correctly classified. These results again highlight that the
discrimination between leg and breast muscle is more evident
than discrimination at species level.
The GA-MLR method that is used below for the analysis of
the FT-IR spectra could not be used on the Raman spectra
because the ratio of the number of objects to variables was too
low. As reported previously this constraint meant that the
algorithm became unstable, and therefore the optimisation of
MLR was not possible in this instance.
48
Analysis of poultry using FT-IR spectroscopy with GA-MLR
Initially cluster analysis was used as described above for the
analysis of the FT-IR spectra from the four muscle types. Very
similar results were seen as described for the Raman spectra
(data not shown) indicating that FT-IR could also be used for
the identification of the muscle type and species from which
the meat had been obtained. The next stage was to ascertain if
there were any spectral features (of those 441 wavenumbers
collected) that could be used for discrimination, rather than
using a full spectral chemometric approach. Therefore GA-
MLR was used, as described above, to differentiate between
(1) different poultry species, (2) different muscle types, and (3)
each of the four meat samples.
It was notable for all GA-MLR models that very few
wavenumbers, of the total 441 possible, were found to be
highly discriminatory illustrating the power of GA for variable
selection prior to MLR. Note that in all of these models one
experiment was used for calibration (evolution) whilst the
other was retained to test the models’ ability to generalise.
That is to say, be predictive for spectra not used in the training
process.
A total of only 8 FT-IR wavenumbers were required by GA-
MLR to discriminate between general muscle type, species and
each of four separate muscle groups from the closely related
muscle foods; the details of these FT-IR bands are shown in
Fig. 1 and summarised in Table 2. In the case of discrimination
of muscle type the three wavenumbers selected were 1413 cm
21
,
1444 cm
21
and 1729 cm
21
and these wavenumbers could be
ascribed to C–N stretch from amides, N–H bend from amides
and CLO stretch from saturated aliphatic aldehyde, respec-
tively. Moreover, the wavenumber 1729 cm
21
was unique for
the discrimination of muscle type. For discrimination at the
species level only two wavenumbers were selected, 1575 cm
21
and 1606 cm
21
, which can be ascribed to a CNH combination
vibration from amide II and NH
2
deformation from amines,
respectively.
For the discrimination of chicken leg muscle from all other
meat samples three wavenumbers were selected, and these were
942 cm
21
, 988 cm
21
and 1606 cm
21
which can be ascribed to
O–H deformation from a carboxyl group, P–O–P stretch from
phosphorous (likely from nucleic acids) and NH
2
deformation
from amines respectively. Moreover, the wavenumber at
Fig. 4 PC-DFA plot constructed from the Raman spectra. 20 PCs
accounting for 98.7% of the explained variance were used in DFA with
a priori knowledge of the muscle and species type (i.e., 4 classes in
total). Upper case letters denote the data from two experiments that
were used to construct the PC-DFA model, whilst lower case letters
denote the test set from an independent experiment. CB + cb = chicken
breast, CL + cl = chicken leg, TB + tb = turkey breast, TL + tl =
turkey leg.
Table 2 Vibrations selected by GA-MLR as being the most
discriminatory for various muscle types and species
Wavenumber/cm
21a
942 988 1382 1413 1444 1575 1606 1729
Muscle type && &
Species &&
Chicken leg && &
Chicken breast && &
Turkey leg &&
Turkey breast &&
a
Marked cells highlight the discriminatory wavenumbers.
1652 |
Analyst
, 2005, 130, 1648–1654 This journal is ßThe Royal Society of Chemistry 2005

988 cm
21
was only selected to discriminate for chicken leg
muscle whilst 1606 cm
21
was also selected to discriminate at
the species level.
In the case of chicken breast muscle three wavenumbers
were selected, 1382 cm
21
, 1413 cm
21
and 1575 cm
21
, ascrib-
able to C–N stretch from amines, C–N stretch from amides,
and CNH combination vibration from amide II respectively.
The wavenumber at 1413 cm
21
was previously selected to
discriminate muscle type and 1575 cm
21
had also been selected
to discriminate at the species level. From the analysis of turkey
leg muscle only two wavenumbers were selected, 942 cm
21
and
1444 cm
21
ascribable to O–H deformation from a carboxyl
group and N–H bend from amides. The wavenumber at
942 cm
21
was previously selected to discriminate chicken leg
muscle and 1444 cm
21
was selected to discriminate at the
species level. As with turkey leg muscle, only two wavenum-
bers were selected to discriminate turkey breast muscle from
all others and these were 1382 cm
21
and 1413 cm
21
, attri-
butable to C–N stretch from amines and C–N stretch from
amides respectively. Finally, the two wavenumbers selected to
discriminate turkey breast muscle were also both selected to
discriminate chicken breast muscle, the only difference
between the 2 muscle types being that an additional wave-
number was selected to discriminate chicken breast muscle at
1575 cm
21
, and significantly, this particular wavenumber was
also selected for discrimination at the species level.
Several of the wavenumbers of interest could be readily
ascribed to proteinaceous groups and indeed this is not
surprising given that the substrate under analysis was muscle
tissue. As previously described in the literature, muscle tissue
contains mostly water (y75%) and protein (18–20%) with the
remainder containing fats, carbohydrates and minerals.
45,49
Therefore, the major group of compounds in muscle tissue
after water are proteinaceous and these can be further sub-
divided into myofibrillar, sarcoplasmic and connective tissue
proteins which constitute approximately 60, 30 and 10% of the
muscle proteins respectively.
45
Two of the discrimination models contained a single
wavenumber particular to that group, such as turkey leg and
more importantly muscle type. The wavenumber assigned to
muscle type at 1729 cm
21
and ascribable to saturated aliphatic
aldehyde, for example, could be as a result of lipid oxidation.
50
Moreover, aldehyde has also previously been shown to be
related to the metabolic type of skeletal muscle
51,52
and its
levels can also vary according to feeding regime, such as the
use of food supplements.
53–55
What was evident was the fact
that all the models contained a combination of either 2 or 3
wavenumbers which were particular to a single muscle group
of a particular species and which distinguished them from
all of the others. An example of the wavenumbers selected by
GA-MLR to discriminate chicken leg from the other muscle
groups is shown in Fig. 5. In this figure the chicken leg samples
are clearly recovered separately from the other three muscle
types, and similar results were seen when each of the
wavenumbers were plotted for the discrimination of each of
the other three meat types (data not shown).
In conclusion, these data clearly demonstrate the utility of
these analytical approaches, based on FT-IR and Raman
spectroscopy, which in combination with appropriate machine
learning-based strategies, provides a rapid, robust and
accurate method for the authentication of closely related
muscle foods. In particular, and as a result of FT-IR analyses
Fig. 5 Pseudo 3-dimensional plot of the intensity of 942 cm
21
plotted against 988 cm
21
and 1606 cm
21
from FT-IR data. Crosses denote data
used in the construction of the GA-MLR model from experiment 1 to discriminate between muscle types from poultry. Circles are data from a
separate second experiment. Other orientations show the same discrimination.
This journal is ßThe Royal Society of Chemistry 2005
Analyst
, 2005, 130, 1648–1654 |1653

by GAs, it was possible to discriminate qualitatively between
all four muscle groups and to find wavenumbers of particular
relevance, between not only the muscle groups of a particular
species, but those wavenumbers of interest which were
responsible for the general discrimination at both the muscle
type and species level. Finally, this is the first time that FT-IR
and Raman spectroscopy have been used successfully for the
discrimination of both closely related poultry species (chicken
and turkey) and for the differentiation of leg from the more
expensive breast muscle. We therefore believe that vibrational
spectroscopies could aid in the identification of specific muscle
groups and also assist in any future quantitative studies
concerning the adulteration of one meat type by another.
Acknowledgements
We are indebted to the Agri-Food and Engineering and
Biological Systems Committees of the UK BBSRC for
supporting this work and we are very grateful to Graham
Price (University of Wales, Aberystwyth) for technical
assistance.
References
1 M. R. Adams and M. O. Moss, Food Microbiology, Royal Society
of Chemistry, Cambridge, 2000.
2 S. Fox, Infect. Med., 2001, 18, 69.
3 J. Pickrell and M. Enserink, Science, 2001, 291, 1677.
4 R. M. Toorop, S. J. Murch and R. O. Ball, Food Res. Int., 1997, 30,
629–636.
5 A. F. Gonzalez-Cordova, A. M. C. de la Barca, M. Cota and
B. Vallejo-Cordoba, Food Sci. Technol. Int., 1998, 4, 257–262.
6 D. R. Martin, J. Chan and J. Y. Chiu, J. Food Prot., 1998, 61,
1686–1690.
7 O. Al-Jowder, M. Defernez, E. K. Kemsley and R. H. Wilson,
J. Agric. Food Chem., 1999, 47, 3210–3218.
8 H. B. Ding and R. J. Xu, J. Agric. Food Chem., 2000, 48,
2193–2198.
9 P. Lingaiah and P. Reddy, J. Food Sci. Technol., 2001, 38, 400–401.
10 R. Goodacre, D. B. Kell and G. Bianchi, Nature, 1992, 359, 594.
11 R. Goodacre, Appl. Spectrosc., 1997, 51, 1144–1153.
12 R. Goodacre, D. Hammond and D. B. Kell, J. Anal. Appl.
Pyrolysis, 1997, 40, 135–158.
13 B. S. Radovic, R. Goodacre and E. Anklam, J. Anal. Appl.
Pyrolysis, 2001, 60, 79–87.
14 J. H. Houben, Fleischwirtschaft, 2002, 82, 133–135.
15 K. D. Hargin, Meat Sci., 1996, 43, 277–289.
16 I. D. Lumley, in Food Authentication, ed. P. R. Ashurst and
M. J. Dennis, Blackie Academic and Professional, London, 1996,
pp. 108–139.
17 O. Al-Jowder, E. K. Kemsley and R. H. Wilson, Food Chem., 1997,
59, 195–201.
18 G. Downey, J. McElhinney and T. Fearn, Appl. Spectrosc., 2000,
54, 894–899.
19 M. E. Flores-Munguia, M. C. Bermudez-Almada and L. Vazquez-
Moreno, J. Muscle Foods, 2000, 11, 319–325.
20 A. K. Lockley and R. G. Bardsley, Trends Food Sci. Technol., 2000,
11, 67–77.
21 U. V. Mandokhot and S. K. Kotwal, J. Food Sci. Technol., 1997,
34, 369–380.
22 S. C. Sheu and Y. H. P. Hsieh, Meat Sci., 1998, 50, 315–326.
23 A. Macedo-Silva, S. F. C. Barbosa, M. G. A. Alkmin, A. J. Vaz,
M. Shimokomaki and A. Tenuta, Meat Sci., 2000, 56, 189–192.
24 A. Skrokki and O. Hormi, Meat Sci., 1994, 38, 497–501.
25 P. M. Reddy, J. Food Sci. Technol., 2001, 38, 32–37.
26 A. A. Khan, H. M. Khan and H. Delincee, Eur. Food Res.
Technol., 2003, 216, 88–92.
27 R. M. Toorop, S. J. Murch and R. O. Ball, Food Res. Int., 1997, 30,
619–627.
28 E. C. Y. Li-Chan, Trends Food Sci. Technol., 1996, 7, 361–370.
29 D. I. Ellis and R. Goodacre, Trends Food Sci. Technol., 2001, 12,
414–424.
30 D. I. Ellis, D. Broadhurst, D. B. Kell, J. J. Rowland and
R. Goodacre, Appl. Environ. Microbiol., 2002, 68, 2822–2828.
31 R. Goodacre, B. S. Radovic and E. Anklam, Appl. Spectrosc.,
2002, 56, 521–527.
32 D. I. Ellis, D. I. Broadhurst and R. Goodacre, Anal. Chim. Acta,
2004, 514, 193–201.
33 W. B. Dunn and D. I. Ellis, Trends Anal. Chem., 2005, 24,
285–294.
34 H. Rannou and G. Downey, Anal. Commun., 1997, 34, 401–404.
35 MAFF, National Food Survey 1999, The Stationary Office,
London, 2000.
36 K. P. J. Williams, G. D. Pitt, B. J. E. Smith, A. Whitley,
D. N. Batchelder and I. P. Hayward, J. Raman Spectrosc., 1994,
25, 131–138.
37A.D.Shaw,N.Kaderbhai,A.Jones,A.M.Woodward,
R. Goodacre, J. J. Rowland and D. B. Kell, Appl. Spectrosc.,
1999, 53, 1419–1428.
38 E. C. Lo´pez-Dı´ez, G. Bianchi and R. Goodacre, J. Agric. Food
Chem., 2003, 51, 6145–6150.
39 R. Goodacre, E. M. Timmins, R. Burton, N. Kaderbhai,
A. M. Woodward, D. B. Kell and P. J. Rooney, Microbiology
(Reading, UK), 1998, 144, 1157–1170.
40 I. T. Jolliffe, Principal Component Analysis, Springer-Verlag,
New York, 1986.
41 H. Wold, in Multivariate Analysis, ed. K. R. Krishnaiah, Academic
Press, New York, 1966, pp. 391–420.
42 B. F. J. Manly, Multivariate Statistical Methods: A Primer,
Chapman and Hall, London, 1994.
43 D. Broadhurst, R. Goodacre, A. Jones, J. J. Rowland and
D. B. Kell, Anal. Chim. Acta, 1997, 348, 71–86.
44 H. E. Johnson, D. Broadhurst, R. Goodacre and A. R. Smith,
Phytochemistry, 2003, 62, 919–928.
45 F. J. G. Schreurs, World’s Poult. Sci. J., 2000, 56, 319–346.
46 R. M. Jarvis and R. Goodacre, Anal. Chem., 2004, 76, 40–47.
47 E. C. Lo´pez-Dı´ez and R. Goodacre, Anal. Chem., 2004, 76,
585–591.
48 A. C. McGovern, D. Broadhurst, J. Taylor, N. Kaderbhai,
M. K. Winson, D. A. Small, J. J. Rowland, D. B. Kell and
R. Goodacre, Biotechnol. Bioeng., 2002, 78, 527–538.
49 R. E. Lawrie, in Meat Science, ed. R. E. Lawrie, Pergamon Press,
London, 1966, pp. 66–113.
50 P. A. Morrissey, J. P. Kerry and K. Galvin, in Freshness and Shelf
Life of Foods, American Chemical Society, Washington, 2003,
pp. 188–200.
51 A. C. Fogerty, F. B. Whitfield, D. Svoronos and G. L. Ford, Int. J.
Food Sci. Technol., 1991, 26, 363–371.
52 C. Alasnier and G. Gandemer, Meat Sci., 1998, 48, 225–235.
53 A. Pfalzgraf, H. Steinhart, M. Kuhn and G. Burgstaller, Fett Wiss.
Technol., 1995, 97, 159–163.
54 M. Du, G. Cherian, P. A. Stitt and D. U. Ahn, Poult. Sci., 2002,
81, 1385–1391.
55 J. S. Elmore, M. M. Campo, M. Enser and D. S. Mottram, J. Agric.
Food Chem., 2002, 50, 1126–1132.
1654 |
Analyst
, 2005, 130, 1648–1654 This journal is ßThe Royal Society of Chemistry 2005