- Email: [email protected]
13C NMR as a tool for authentication of different gadoid fish species with emphasis on phospholipid profiles
Food Chemistry 121 (2010) 608–615
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Analytical Methods 13
C NMR as a tool for authentication of different gadoid fish species with emphasis on phospholipid profiles Inger B. Standal a,b,*, David E. Axelson c, Marit Aursand a a
SINTEF Fisheries and Aquaculture, N-7465 Trondheim, Norway Department of Biotechnology, Norwegian University of Natural Science and Technology, N-7491 Trondheim, Norway c MRi_Consulting, 8 Wilmot St. Kingston, Ontario, Canada K7L 4V1 b
a r t i c l e
i n f o
Article history: Received 16 December 2008 Received in revised form 30 October 2009 Accepted 28 December 2009
Keywords: 13 C NMR spectroscopy Authentication Phospholipids Multivariate data analyses Stereospecific Positional distribution PCA BBN Atlantic cod North-east arctic cod Norwegian coastal cod Haddock Saithe Pollack
a b s t r a c t The aim of this study was to evaluate if phospholipid profiles obtained by 13C nuclear magnetic resonance (NMR) spectroscopy is characteristic enough to separate species of lean gadoid fish. 13C NMR data were obtained from muscle lipids of five categories of lean gadoid fish, namely, north-east arctic cod and Norwegian coastal cod (Gadus morhua), haddock (Melanogrammus aeglifinus), saithe (Pollachius virens), and pollack (P. pollachius). A total of 27 fish caught at the same location on the Norwegian coast in the traditional fishing season (March/April) in 2006 were analysed. The sn-2 position specificity of 22:6n-3 (docosahexaenoic acid, DHA) in phosphatidyl choline (PC) and phosphatidyl ethanolamine (PE) for the different species/stocks were investigated, and the full 13C NMR spectra applied in multivariate analysis. Stereospecific distribution calculations showed significant differences among species in the distribution of 22:6n-3 in PC and PE, and the pollack group displayed the lowest values for 22:6n-3 in sn-2 position, both in PC and PE. This first screening showed that by using the 13C NMR fingerprint of muscle lipids, linear discriminant analysis gave a correct classification rate of 78% according to the five categories of lean gadoid fish, while successful classification (100%) was achieved with Bayesian belief networks (BBN) predictions. Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction Gadoid species, such as Atlantic cod, saithe, haddock and pollack, constitute the most important species in Norwegian fisheries. In 2007 550,000 tonnes of these fish species were landed from Norwegian wessels. The catches of Atlantic cod and saithe are approximately similar (220,000 tonnes), but the catch values of Atlantic cod (north-east arctic cod) and saithe is 31% and 10% respectively (Fiskeridirektoratet, 2007). EU directives have introduced labelling regulations which requires that fishery and aquaculture products should be labelled with information such as: species, geographical origin, and production method of fish (i.e. wild/farmed) (EC, 2001). However, differences in quality and price between fish of different species and origin, may lead to falsification and mislabelling, and there is a need for methods able to verify
* Corresponding author. Address: Department of Biotechnology, Norwegian University of Natural Science and Technology, N-7491 Trondheim, Norway. Tel.: +47 98222468; fax: +47 93270701. E-mail address: [email protected] (I.B. Standal). 0308-8146/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.foodchem.2009.12.074
that the traceability information is correct, to protect consumer rights and to prevent illegal capture. Traditional methods for species identification of fish are DNA or protein analyses (Martinez et al., 2003), however it would be convenient to use additional and complementary techniques for processed products. Analysis of lipids is a potential tool for authentication of fish and marine oils (Aursand, Standal, & Axelson, 2007; Aursand, Standal, Praël, McEvoy, & Axelson, 2009; Hidalgo & Zamora, 2003). Triacylglycerol (TAG) fatty acid composition in muscle of fatty fish, such as salmon, and in livers of gadoid fish, such as cod, reflects the diet (dos Santos, Burkow, & Jobling, 1993; Lie, 1991), which makes it possible to discriminate between wild and farmed fish from lipid analysis (Aursand, Mabon, & Martin, 2000; Aursand et al., 2009; Standal, Praël, McEvoy, Axelson, & Aursand 2008). However, fatty acid composition of fish is a net result of a wide range of factors, including, lipid metabolism, season, age, size and stage of sexual maturity and environmental factors (Sargent, Bell, McEvoy, Tocher, & Estevez, 1999). It has also been shown that certain fatty acids seem to be maintained within limits which depend on the species (Aursand, Jørgensen, & Grasdalen, 1995). Mus-
609
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
cle of the lean marine fish species consists primarily of phospholipids (PLs), and even though polar lipids are affected by diet to some degree (Mørkøre, Netteberg, Johnson, & Pickova, 2007), they are less influenced by diet than TAGs (Lie, Hemre, & Lambertsen 1992; Nanton, Lall, & McNiven, 2001). Therefore, when studying species or stock differences, PL analysis is recommended (Joensen, Steingrund, Fjallstein, & Grahl-Nielsen, 2000). Analyses on eggs from various cod stocks, suggested that fatty acid compositions of PLs were more dependent on stocks then on diet (Pickova, Dutta, Larsson, & Kiessling, 1997). Species of Atlantic tuna have been distinguished from PL–fatty acid profiles of white muscle (Medina, Aubourg, & Pérez Martín, 1997). The standard analysis of fatty acids is by gas chromatography (GC), but the use of high resolution (HR) 13C nuclear magnetic resonance (NMR) spectroscopy in the analysis of lipids is increasing. 13 C NMR is a multicomponent technique, which gives, in addition to fatty acid composition as GC, information on lipid classes and sn-2 position specificity of fatty acids in TAGs, which are characteristic for certain species (Aursand et al., 1995), and PLs (Falch, Størset, & Aursand, 2006). Previous analyses of marine oils showed that one could easily distinguish oils of different species according to their TAG profile obtained by 13C NMR (Aursand et al., 1995; Standal, Axelson, & Aursand, 2009). 13 C NMR studies on marine PLs include analysis on cod gonads (Falch, Størseth, & Aursand, 2007; Falch et al., 2006), and tuna muscle PLs (Medina & Sacchi, 1994; Sacchi et al., 1993). However, to the authors’ knowledge, 13C NMR studies of muscle lipids extracted from different species and stocks of lean gadoid fish, have not been published previously. The aim of this study was to evaluate if 13C NMR phospholipid profiles are characteristic enough to discriminate fish species and stocks. For the study five categories of lean gadoid fish were selected, namely: two stocks of Atlantic cod (Gadus morhua L.) (north-east arctic cod and Norwegian coastal cod), and the species haddock (Melanogrammus aeglifinus), saithe (Pollachius virens), and pollack (P. pollachius). The 13C NMR spectra were investigated to observe any differences in stereospecific distribution of the fatty acid 22:6n-3 in phosphatidyl choline (PC) and phosphatidyl ethanolamine (PE) among the different groups analysed. The full 13C NMR fingerprints were applied in multivariate analysis to observe groupings, to classify samples according to species/stocks, and to determine important variables in the classification. 2. Materials and methods 2.1. Materials The following species and stocks of lean fish were caught outside the coast of Vikna, Nord-Trondelag, Norway during March/ April 2006 (in the traditional fishing season): north-east arctic cod (AC, n = 6) and Norwegian coastal cod (CC, n = 6) (G. morhua L.), haddock (H, n = 6) (M. aeglefinus), saithe (S, n = 5) (P. virens), pollack (P, n = 4) (P. pollachius). The fish was frozen within 24 h by the fisherman and transported to the laboratory. In total, 27 fish were analysed. The average round weight of the different fish categories (with standard deviation) are north-east arctic cod: 2.9 ± 0.6 kg, coastal cod: 2.2 ± 0.3 kg, haddock: 0.7 ± 0.2 kg, saithe: 2.7 ± 1.2 kg and pollack: 4.3 ± 1.1 kg.
2.3.
13
C NMR
Approximately 90 mg of the oil sample was transferred to 5 mm NMR tubes and diluted with 0.6 mL deuteriated chloroform (CDCl3, 99.8% purity, Isotec Inc., Matheson). The 13C NMR spectra were run semi-quantitatively with a high number of scans, to achieve sufficient signal to noise (S/N) ratio to evaluate the stereospecific distribution of fatty acids in PLs. Comparisons between intensities of resonances within the same spectra should be performed with care in semi-quantitative 13C NMR spectra, however, comparison of signals among the different spectra is possible. Previous studies have shown that a semi-quantitative approach can be applied in positional distribution measurements, since the nuclear overhauser effect (NOE) effect is similar for carbonyl-carbons and that the T1 values of these carbons does not vary much according to the position of the fatty acids in the glycerol molecule (Aursand et al., 1995; Wollenberg, 1990). 13C NMR spectra were obtained on a Bruker Avance DMX 600 instrument (Bruker BioSpin GmbH, Rheinstetten, Germany) operating at 150.9 MHz for carbons, using a 5 mm BBO probe at 298 K. Power gated decoupling was applied. The following acquisition parameters were used: time domain 64k, pulse width 60°, sweep-width 200.8 ppm, acquisition time 1.08 s, relaxation delay 0.5 s, number of scans 16k. Zero filling and exponential line broadening (0.15 Hz) were applied before Fourier transform. The chemical shift scale is referred indirectly to TMS by the triplet of CDCl3 at 77.11 ppm. 2.4. Peak picking of full
13
C NMR spectra
Peak positions and intensities were obtained for resonances >0.5% of the maximum peak intensity within each spectrum. The resulting peak list was exported for manual alignment due to small variations in chemical shifts among samples. The solvent resonances were removed, before the data matrix (254 variables for the 27 samples investigated) were exported for multivariate analysis. 2.5. Calculation of the sn-2 position specificity of 22:6n-3 in PC and PE Peakfitting was applied to the carbonyl region of the 13C NMR spectra, to facilitate integration of peaks arising from 22:6n-3 in sn-1,3 and sn-2 position of PC and PE and to reveal hidden peaks (in particular 22:6n-3 in sn-2 position of PE consisted of more than one peak). The 13C NMR data were processed in xwinnmr as previously described, and the full spectra (1r file) were converted into ASCII files. Data in the carbonyl region (174–172 ppm) were selected (681 data points) and imported to PeakFit 4.12 (SeaSolve Software Inc, San Jose, CA). The AutoFit Residuals method was chosen as peak-fitting method. This procedure initially places peaks by finding local maxima in a smoothed data stream. Hidden peaks are than optionally added where peaks in the residuals occur (Peakfit 4.12). Peaks were fitted, assuming Lorentzian lineshape, with a linear baseline substracted prior to fitting. The area of the peaks arising from sn-1 and sn-2 22:6n-3 in PC and PE were registered, and the relative distribution of 22:6n-3 in sn-2 position calculated. Comparisons between means for the five fish categories were performed by one-way analysis of variance (ANOVA) with a significance level of 0.01 (Excel, Microsoft XP). 2.6. Multivariate analysis
2.2. Lipid extraction Lipid was extracted from white fish muscle under the back dorsal fin according the Bligh and Dyer method (1959). Before analysing the lipid extract by NMR, parts of the chloroform phase were removed by evaporation.
The 13C NMR data were analysed by the multivariate methods principal component analysis (PCA) (Jolliffe, 1986; Wold, Esbensen, & Geladi, 1987), linear discriminant analysis (LDA) and Bayesian belief network (BBN) (Heckerman, Geiger, & Chickering, 1995; Pearl, 1988) to observe groupings and to test if it is possible to
610
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
discriminate fish of different species (cod, haddock, saithe, pollack) and stocks (north-east arctic cod, Norwegian coastal cod). PCA was applied as an unsupervised multivariate technique. PCA has frequently been applied to spectral data for dimensionality reduction, to identify outliers and to classify samples. In PCA the original variables are transformed into new, uncorrelated variables called principal components, which retains as much as possible of the information present in the original data. Each principal component (PC) is a linear combination of the original variables. The scores of a subset of the principal components can be used in subsequent multivariate analysis. LDA is a supervised pattern recognition technique, which is based on the assumption that samples of the same group are more similar than samples belonging to different groups. The technique seeks to find a linear transformation by maximising the betweenclass variance and minimising the within-class variance. LDA has been widely used for pattern recognition and data analysis. In situations where the number of variables exceeds the number of ob-
AC1
jects, the principal components can be used as variables in the linear discriminant analysis (Kim, Kim, & Bang, 2003). BBN analyses have been used extensively in the area of medical diagnostics and bioinformatics, and recent studies have also shown the potential of this technique for analyses of HR NMR data (Martinez et al., 2009). A Bayesian network consists of: (1) nodes which represent the random variables, where each node has states (a set of probable values for each variable), (2) directed edges (arrows) which connect the nodes (represent dependencies, absence of arrows indicate independence), (3) a conditional probability table which is associated with each node (prior probability), and (4) a directed acyclic graph where the graph represents independence relationships between variables. The most important advantages for this work are that BBNs calculate explicit probabilities for a hypothesis, can be used as a classifier and can find the variables with the most impact on the classification. A more thorough description of the BBN method is given in (Axelson, Standal, Martinez, & Aursand, 2009).
PC (sn-3)
PC (CH2O)
PE/PC (sn-1)
PL (sn-2) TAG (sn-1,3)
PE (CH2O)
TAG (sn-2)
70
69
68
67
66
65
64
63
ppm
70
69
68
67
66
65
64
63
ppm
70
69
68
67
66
65
64
63
ppm
70
69
68
67
66
65
64
63
ppm
70
69
68
67
66
65
64
63
ppm
CC1
H1
S5
P4
Fig. 1. Glycerol region of 13C NMR spectra of total lipids extracted from white muscle of north-east arctic cod (AC), Norwegian coastal cod (CC), haddock (H), saithe (S) and pollack (P).
611
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
The NMR data were normalised to an area of 100, mean centred and autoscaled prior to multivariate analyses. PCA was carried out (Minitab15, Minitab Inc.) by using the 254 chemical shift intensities as input data (with chemical shifts from solvents excluded in analysis). In the LDA (Minitab, version 15), the sample scores on PC components 1–3 in the previously mentioned PCA calculations, were used as input variables. The analyses were performed with sample grouping 1–5 according to species/stock (north-east arctic cod, Norwegian coastal cod, haddock, saithe, pollack). Cross validation (CV) was used as validation method when LDA models were developed. In the leave one out (LOO) CV each sample is removed one at a time, the classification function is recalculated using the remaining data, and the omitted observation is classified. BBN predictions (Netica v4.02, Norsys Sofware Corp. Vancouver B.C., Canada) were made for 18 samples that were randomly se-
lected in each group for the training set, while nine were held back for the validation calculations. The variables with most impact on the classification were identified from the BBN analysis, and corresponding NMR peaks were assigned according to previous analysis on marine PLs (Falch et al., 2006, 2007).
3. Results and discussion 3.1.
173.6
173.4
22:6n-3 (sn-2) PC
20:5n-3 and 20:4n-6 (sn-2) PC
MUFA/SFA (sn-1,3) MUFA/SFA (sn-2) TAG PL
173.8
C NMR spectra
Assignments of the 13C NMR spectra of total lipids extracted from fish muscle in this study were made according to previous studies on marine PLs (Falch et al., 2006, 2007). The glycerol and polar-headgroup region (61–71 ppm, Fig. 1) gives information
MUFA/SFA (sn-1) PL
AC1
13
173.2
22:6n-3 (sn-1) 22:6n-3 PC (sn-1) PE
173.0
22:6n-3 (sn-2) PE
172.8
172.6
172.4
172.2
ppm
MUFA/SFA (sn-2) TAG and 20:5n-3 (sn-2) PE
CC1
173.8
173.6
173.4
173.2
173.0
172.8
172.6
172.4
172.2
ppm
173.8
173.6
173.4
173.2
173.0
172.8
172.6
172.4
172.2
ppm
173.8
173.6
173.4
173.2
173.0
172.8
172.6
172.4
172.2
ppm
173.8
173.6
173.4
173.2
173.0
172.8
172.6
172.4
172.2
ppm
H1
S5
P4
Fig. 2. Carbonyl region of 13C NMR spectra of total lipids extracted from white muscle of north-east arctic cod (AC), Norwegian coastal cod (CC), haddock (H), saithe (S) and pollack (P).
612
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
from cholesterol could be found in other parts of the spectra (Falch et al., 2006). Lipids from white muscle of lean gadoid fish contain approximately about 73–78% PLs, 3–9% TAGs, and 6–9% sterols (dos Santos et al., 1993). Even though the glycerol- and polar-headgroup region (Fig. 1) shows characteristics according to stereospecific conformation, the carbonyl region is better suited for the purpose of obtaining data on the stereospecific distribution of fatty acids (Fig. 2). From Fig. 2, one can see variation in relative peak intensities among the five fish categories. Most pronounced is the variation in the peak intensity arising from sn-2 20:5n-3/20:4n-6 in PC at 172.95 ppm. However, this variation in sn-2 20:5n-3/20:4n-6 content is presumably not species-specific, as large intra-specific differences of the same fish species (e.g. 5 coastal cods) was observed (results not shown). The fish in this study were caught in the traditional fishing season of these species, however this is identical with the spawning time, and the intra-specific differences in lipid composition may be due to variations in reproductive status, but also life stage, size, sex, food availability and environmental factors may influence fatty acid composition (Sargent et al., 1999).
Table 1 Average values on the sn-2 position specificity of 22:6n-3 in PC and PE (mole%). AC: north-east arctic cod, CC: Norwegian coastal cod, H: haddock, S: saithe, P: Pollack.
AC CC H S P
n=6 n=6 n=6 n=5 n=4
DHA sn-2 PC
DHA sn-2 PE
84 ± 3 78 ± 3 85 ± 3 80 ± 3 77 ± 3
76 ± 3 71 ± 1 71 ± 5 76 ± 5 65 ± 4
about lipid classes present in the sample. Signals from the principal PL species in cod muscle, PC and PE, but also traces of TAGs could be seen in the spectra (Fig. 1). Even though there were minor differences in the level of triacylglycerols (TAGs) among the samples (as one can also observe for the five samples in Fig. 1), the level of triacylglycerols was considered too low to have a significant impact on the fatty acid composition. As an example; only one sample (of saithe) showed relatively high levels of TAGs (approximately twice the level of sample S5 in Fig. 1), but still displayed similar fatty acid composition as the other saithes. Several resonances
(a)
AC3
10 AC1 AC5 AC2 AC6 AC4
PC2
5
0
CC5 S1 CC6 CC2 S2 CC4 S4
S5
H4
H5 CC1
P2
-5
H6
H2 CC3
H1
H3
P3 P4
P1 S3
-10 -10
(b)
0
10 PC1
20
30
10 CC4 CC2
5
H4
CC5 P2 P3
CC6 H3
P1
AC6 AC2 AC4AC3 AC5 AC1
S5
H5
H2 CC1
PC3
0
CC3
P4
H1
H6 S4
-5
S3 S2
-10 S1
-15 -10
0
10 PC1
20
30
Fig. 3. PC1 vs. PC2 (a) and PC1 vs. PC3 (b) score plot from NMR data for the five different groups of fish. The three PCs explain 36%, 9% and 9% of the variance in the dataset respectively. AC: north-east arctic cod, CC: Norwegian coastal cod, H: haddock, S: saithe, P: pollack (254 chemical shifts as input variables).
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
3.2. Stereospecific calculations As previously mentioned, the stereospecific distributions of TAGs in fish lipids are quite specific for certain species (Aursand et al., 1995). To evaluate if similar differences could be found in marine PLs, the sn-2 position specificity of 22:6n-3 in PC and PE was calculated from the 13C NMR spectra. Well resolved peaks from 22:6n-3 in sn-2 and sn-1 position in PE and PC appeared in the carbonyl region of the 13C NMR spectra (Fig. 2). This allowed the calculation of the stereospecific distribution of this fatty acid in PC and PE. Average values on the sn-2 position specificity of 22:6n-3 for each species/stocks are given in Table 1. The results showed that 22:6n-3 is mainly esterified to sn-2 of PC in all species, with average values between 77% and 85% for the five categories. Also in PE the sn-2 position was the preferred position for 22:6n-3, with average values for the five groups ranging from 65–76%. These results are in accordance with previous studies on PLs in cod milt and roe, which showed that the n3 polyunsaturated fatty acids (PUFAs) were primarly esterified in the sn-2 position of PC and PE (Bell & Dick, 1991; Falch et al., 2006). In TAGs of cod liver oil, DHA was concentrated in sn-2 posi-
613
tion with 74.4%, while EPA was nearly randomly distributed with 39.7% (Aursand et al., 1995). The pollack group displayed the lowest values of 22:6n-3 in sn2 position both for PE and PC. ANOVA calculations showed that the ratio of DHA in sn-2 PC was significantly lower for the pollacks than for the north-east arctic cods and the haddocks. Also the coastal cod group displayed lower mean value than the haddocks for DHA in sn-2 PC. For PE, the ratio of sn-2 22:6n-3 of was significantly lower for the pollack group than for north-east arctic cods, coastal cods and the saithes. Further studies are needed to conclude whether these differences, are due to genetic differences, or if stereospecific distribution of fatty acids is influenced by factors such as diet, sex and reproductive status. The relatively low values on 22:6n-3 in sn-2 position of PE and PC in pollacks may also be attributed to the generally higher level of 22:6n-3 in these fish (and thereby possibly higher level of 22:6n-3 in sn-1 position) compared to the other fish in this study. A weak linear correlation (negative) were found between DHA levels and DHA in sn-2 positions of PC ( 0.43) and PE ( 0.34). However, to which extent dietary fatty acid composition influence the stereospecific distribution of fatty acids remains to be investigated.
Fig. 4. Distributions of the eight most important variables in the BBN analysis for the classes AC (top) and CC (bottom). Horizontal bars denote relative percentage of values of the given variable (chemical shift, ppm).Values at the bottom are the mean and standard deviation for the chemical shift intensities. AC: north-east arctic cod, CC: Norwegian coastal cod, H: haddock, S: saithe, P: pollack.
614
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
29.29; 29.31; 29.32
34.5
34.0 128.6
28.04
28.14
33.96
29.11
ppm
29.0
28.5
ppm
128.02 128.27 70.77 127.54
128.5 Fig. 5. The
128.0
ppm
71.5
71.0
70.5
ppm
13
C NMR spectrum of north-east arctic cod (CC) with some of the most important variables in the BBN classification highlighted.
3.3. Multivariate analysis on
13
C NMR data
PCA was applied to the NMR data to observe groupings and to interpret differences among groups. The PCA was carried out with 254 chemical shifts as input variables. The score plot of PC1 vs. PC2 is shown in Fig. 3a. Samples with similar PL profiles will be closely located in this score plot. The first two principal components account for 36% and 9% of the variance in the dataset. Clustering of fish of the different groups can be observed; the north-east arctic cod (AC) samples are well separated from the other samples in this score plot, and the pollack (P) samples are also grouped together. However, the coastal cod (CC) samples are positioned far away from each other, and overlaps with the saithes (S) and the haddocks (H). However, also PC3 (explaining 9% of the variance) contains valuable information when it comes to separating the five different classes from NMR data. In this PCA plot, also the saithe (S) samples are closely positioned (Fig. 3b). North-east arctic cod migrates from the feeding areas in the Barents Sea to the spawning areas along the Norwegian coast in March/April, in contrast to the coastal cod, which are quite stationary. Coastal cod is listed as an endangered species on the ‘‘red list” of Norwegian Biodiversity Information Centre (www.artsdatabanken.no). The fact that the north-east arctic cod samples are completely separated from the Norwegian coastal cods in Fig. 3a, indicates that there are differences in lipid composition between these two stocks of Atlantic cod, which may arise from dietary or metabolic differences. The two stocks have previously been distinguished from otholite-, haemoglobin-, and gene analysis (Sarvas & Fevolden, 2005). For a more quantitative result on the classification power of the dataset, LDA and BBN predictions were applied. The three first scores from the previous PCA were used as input variables in the LDA. LDA with the 3PCs obtained from NMR data gave 21/27 correct classifications (78% correctly classified). The BBN approach also allows one to rank the variables according to their relevance to the classification. In the BBN analysis, 100% correct classifications were achieved with the 17 most important chemical shifts as variables. The 17 variables with the most impact on the classification were determined to be: 172.81, 26.56, 33.96, 29.31, 25.61, 29.29, 128.02, 29.14, 128.60, 70.77, 29.32, 28.04, 128.27, 29.11, 25.32, 127.54 ppm. In a subsequent reduction of variables, only one sample was misclassified with eight chemical shifts as variables (one of the coastal cod samples classified as north-east arctic cod in the validation dataset).
A schematic representation of BBN for the classification is given in Fig. 4 (example north-east arctic cod (AC) and Norwegian coastal cod (CC) groups). Horizontal bars denote relative percentage of values of the eight variables with the highest impact in the classification for samples of the indicated class (in this example: AC and CC). Values at the bottom are the mean and standard deviation for the given chemical shift. Assignment of important chemical shifts in the classification of the five fish categories were made according to Falch et al., 2006. The peak at 172.81 ppm (Fig. 2) arise from carbonyl-carbon of 22:6n-3 in sn-1 PC. Peaks around 33.96 ppm stem from 22:6n-3 sn-2 PL, while the singlet at 28.04 ppm arise from carbon number 19 of cholesterol (Fig. 5). However, other peaks were more difficult to identify. In example, the peaks at 29.32, 29.31, and 29.29 ppm (Fig. 5) arise from carbon atoms of various saturated, mono-and di unsaturated-fatty acids.
3.4. Conclusion This first screening indicates that it is possible to discriminate amongst species and stocks of lean gadoid fish based on lipid profiling of muscle lipids, which are mainly composed of phospholipids. Lipid profiles from 13C NMR gave 100% classification of the five categories of fish by the BBN approach. In a further evaluation of the reliability of this approach for authentication purposes, it would be important to map the intra-specific variation in the phospholipid profiles due to factors such as diet, year of catch, season, reproductive status, sex and geographical origin. If the profiles are characteristic of species regardless of such factors, it would, as for most authentication methods, be necessary to establish an extensive database with reference material covering this natural variation. 13 C NMR is a less sensitive technique than GC, but has the advantage that it can be applied non-destructively on intact lipids. Additional information is obtained from 13C NMR than from GC, such as information on lipid classes and the stereospecific distribution of fatty acids, which might be relevant for authentication purposes. Results on sn-2 position specificity of 22:6n-3 in PC and PE did show that there were significant differences between groups of fish, however, further studies are needed to conclude whether these differences can be attributed as genetic differences, and to which extent the stereospecific distribution of fatty acids is influenced by factors such as diet, sex and reproductive status.
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
Acknowledgements The Norwegian Research Council is gratefully acknowledged by the financial support to NFR-project no. 146932/130. Thanks also to Terje Breivik for collection of the fish. References Aursand, M., Jørgensen, L., & Grasdalen, H. (1995). Positional distribution of x3 fatty acids in marine lipid triacylglycerols by high-resolution 13C nuclear magnetic resonance spectroscopy. Journal of the American Oil Chemists’ Society, 72, 293–297. Aursand, M., Mabon, F., & Martin, G. J. (2000). Characterization of farmed and wild salmon (Salmo salar) by a combined use of compositional and isotopic analysis. Journal of the American Oil Chemists’ Society, 77, 659–666. Aursand, M., Standal, I. B., & Axelson, D. E. (2007). High-resolution 13C NMR spectroscopy pattern recognition of fish oil capsules. Journal of Agricultural and Food Chemistry, 55, 38–47. Aursand, M., Standal, I. B., Praël, A., McEvoy, L., & Axelson, D. E. (2009). 13C NMR pattern recognition techniques for the classification of Atlantic salmon (Salmo salar L.) according to their wild, farmed and geographical origin. Journal of Agricultural and Food Chemistry, 57, 3444–3451. Axelson, D. E., Standal, I. B., Martinez, I., & Aursand, M. (2009). Classification of wild and farmed salmon using Bayesian belief networks and gas chromatography derived fatty acid distributions. Journal of Agricultural and Food Chemistry, 57, 7634–7639. Bell, M. V., & Dick, J. R. (1991). Molecular-species composition of the major diacyl glycerophospholipids from muscle, liver, retina and brain of cod (Gadus morhua). Lipids, 26, 565–573. Bligh, E. G., & Dyer, W. J. (1959). A rapid method of total lipid extraction and purification. Canadian Journal of Biochemistry and Physiology, 37, 911–917. dos Santos, J., Burkow, I. C., & Jobling, M. (1993). Patterns of growth and lipid deposition in cod (Gadus morhua L.) fed natural prey and fish-based feeds. Aquaculture, 110, 173–189. EC Commission Regulation No. 2065/2001 of October 2001 laying down detailed rules for the application of Council Regulation (EC) No. 104/2000 as regards informing consumers about fishery and aquaculture products. Falch, E., Størset, T., & Aursand, M. (2006). Multi-component analysis of marine lipids in fish gonads with emphasis on marine phospholipids using high resolution NMR spectroscopy. Chemistry and Physics of Lipids, 144, 4–16. Falch, E., Størseth, T. R., & Aursand, M. (2007). High resolution NMR for studying lipid deterioration in cod (Gadus morhua) gonads. Chemistry and Physics of Lipids, 147, 46–57. Fiskeridirektoratet (2007).. Heckerman, D., Geiger, D., & Chickering, D. (1995). Learning Bayesian networks: The combination of knowledge and statistical data. Machine Learning, 20, 197–243. Hidalgo, J., & Zamora, R. (2003). Edible oil analysis by high-resolution nuclear magnetic resonance spectroscopy: Recent advances and future perspective. Trends in Food Science and Technology, 14, 499–506. Joensen, H., Steingrund, P., Fjallstein, I., & Grahl-Nielsen, O. (2000). Discrimination between two reared stocks of cod (Gadus morhua) from the Faroe Islands by chemometry of the fatty acid composition in the heart tissue. Marine Biology, 136, 573–580.
615
Jolliffe, I. T. (1986). Principal component analysis. Berlin: Springer-Verlag. Kim, H. C., Kim, D., & Bang, S. Y. (2003). Extensions of LDA by PCA mixture model and class-wise features. Pattern Recognition, 36, 1095–1105. Lie, Ø. (1991). Studies on digestion, deposition and fatty acid composition of lipids in cod (Gadus morhua). PhD thesis, University of Bergen, Bergen, Norway. Lie, Ø., Hemre, G. I., & Lambertsen, G. (1992). Influence of dietary fatty acids on the glycerophospholipid composition in organs of cod (Gadus morhua). Lipids, 27, 770–775. Martinez, I., Aursand, M., Erikson, U., Singstad, T. E., Veliuylin, E., & Van der Zwaag, C. (2003). Destructive and non-destructive analytical techniques for authentication and composition analyses of foodstuffs. Trends in Food Science and Technology, 14, 489–498. Martinez, I., Standal, I. B., Axelson, D. E., Finstad, B., & Aursand, M. (2009). Identification of the farm origin of salmon by fatty acid and HR 13C NMR profiling. Food Chemistry, 116, 766–773. Medina, I., Aubourg, P., & Pérez Martín, R. (1997). Species differentiation by multivariate analysis of phospholipids from canned Atlantic tuna. Journal of Agricultural and Food Chemistry, 45, 1997–2499. Medina, I., & Sacchi, R. (1994). Acyl stereospecific analysis of tuna phospholipids via high resolution 13C NMR spectroscopy. Chemistry and Physics of Lipids, 70, 53–61. Mørkøre, T., Netteberg, C., Johnson, L., & Pickova, J. (2007). Impact of dietary oil source on product quality of farmed Atlantic cod, Gadus morhua. Aquaculture, 267, 236–247. Nanton, D. A., Lall, S. P., & McNiven, M. A. (2001). Effects of dietary lipid level on liver and muscle lipid deposition in juvenile haddock, Melanogrammus aeglefinus L. Aquaculture Research, 32, 225–234. Norwegian Biodiversity Information Centre.. Pearl, J. (1988). Probabilistic reasoning for intelligent systems. San Francisco: Morgan Kaufmann. Pickova, J., Dutta, P. C., Larsson, P. O., & Kiessling, A. (1997). Early embryonic cleavage pattern, hatching success, and egg-lipid fatty acid composition: Comparison between two cod (Gadus morhua) stocks. Canadian Journal of Fisheries and Aquatic Science, 54, 2410–2416. Sacchi, R., Medina, I., Aubourg, S. P., Giudicianni, I., Paolillo, L., & Addeo, F. (1993). Quantitative high-resolution 13C NMR analysis of lipids extracted from the white muscle of Atlantic Tuna (Thunnus alalunga). Journal of Agricultural and Food Chemistry, 41, 1247–1253. Sargent, J., Bell, G., McEvoy, L., Tocher, D., & Estevez, A. (1999). Recent developments in the essential fatty acid nutrition of fish. Aquaculture, 177, 191–199. Sarvas, T., & Fevolden, S. E. (2005). Pantophysin Pan I locus divergence between inshore v. Offshore and northern v. southern populations of Atlantic cod in the north-east Atlantic. Journal of Fish Biology, 67, 444–469. Standal, I. B., Axelson, D. E., & Aursand, M. (2009). Differentiation of fish oils according to species by 13C-NMR regiospecific analyses of triacyglycerols. Journal of the American Oil Chemists’ Society, 86, 401–407. Standal, I. B., Praël, A., McEvoy, L., Axelson, D. E., & Aursand, M. (2008). Discrimination of cod liver oil according to wild/farmed and geographical origins by GC and 13C NMR. Journal of the American Oil Chemists’ Society, 85, 105–112. Wold, S., Esbensen, K., & Geladi, P. (1987). Principal component analysis. Chemometrics and Intelligent Laboratory Systems, 2, 37–52. Wollenberg, K. F. (1990). Quantitative high resolution 13C nuclear magnetic resonance of the olefinic and carbonyl carbons of edible vegetable oils. Journal of the American Oil Chemists’ Society, 67, 487–494.
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Analytical Methods 13
C NMR as a tool for authentication of different gadoid fish species with emphasis on phospholipid profiles Inger B. Standal a,b,*, David E. Axelson c, Marit Aursand a a
SINTEF Fisheries and Aquaculture, N-7465 Trondheim, Norway Department of Biotechnology, Norwegian University of Natural Science and Technology, N-7491 Trondheim, Norway c MRi_Consulting, 8 Wilmot St. Kingston, Ontario, Canada K7L 4V1 b
a r t i c l e
i n f o
Article history: Received 16 December 2008 Received in revised form 30 October 2009 Accepted 28 December 2009
Keywords: 13 C NMR spectroscopy Authentication Phospholipids Multivariate data analyses Stereospecific Positional distribution PCA BBN Atlantic cod North-east arctic cod Norwegian coastal cod Haddock Saithe Pollack
a b s t r a c t The aim of this study was to evaluate if phospholipid profiles obtained by 13C nuclear magnetic resonance (NMR) spectroscopy is characteristic enough to separate species of lean gadoid fish. 13C NMR data were obtained from muscle lipids of five categories of lean gadoid fish, namely, north-east arctic cod and Norwegian coastal cod (Gadus morhua), haddock (Melanogrammus aeglifinus), saithe (Pollachius virens), and pollack (P. pollachius). A total of 27 fish caught at the same location on the Norwegian coast in the traditional fishing season (March/April) in 2006 were analysed. The sn-2 position specificity of 22:6n-3 (docosahexaenoic acid, DHA) in phosphatidyl choline (PC) and phosphatidyl ethanolamine (PE) for the different species/stocks were investigated, and the full 13C NMR spectra applied in multivariate analysis. Stereospecific distribution calculations showed significant differences among species in the distribution of 22:6n-3 in PC and PE, and the pollack group displayed the lowest values for 22:6n-3 in sn-2 position, both in PC and PE. This first screening showed that by using the 13C NMR fingerprint of muscle lipids, linear discriminant analysis gave a correct classification rate of 78% according to the five categories of lean gadoid fish, while successful classification (100%) was achieved with Bayesian belief networks (BBN) predictions. Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction Gadoid species, such as Atlantic cod, saithe, haddock and pollack, constitute the most important species in Norwegian fisheries. In 2007 550,000 tonnes of these fish species were landed from Norwegian wessels. The catches of Atlantic cod and saithe are approximately similar (220,000 tonnes), but the catch values of Atlantic cod (north-east arctic cod) and saithe is 31% and 10% respectively (Fiskeridirektoratet, 2007). EU directives have introduced labelling regulations which requires that fishery and aquaculture products should be labelled with information such as: species, geographical origin, and production method of fish (i.e. wild/farmed) (EC, 2001). However, differences in quality and price between fish of different species and origin, may lead to falsification and mislabelling, and there is a need for methods able to verify
* Corresponding author. Address: Department of Biotechnology, Norwegian University of Natural Science and Technology, N-7491 Trondheim, Norway. Tel.: +47 98222468; fax: +47 93270701. E-mail address: [email protected] (I.B. Standal). 0308-8146/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.foodchem.2009.12.074
that the traceability information is correct, to protect consumer rights and to prevent illegal capture. Traditional methods for species identification of fish are DNA or protein analyses (Martinez et al., 2003), however it would be convenient to use additional and complementary techniques for processed products. Analysis of lipids is a potential tool for authentication of fish and marine oils (Aursand, Standal, & Axelson, 2007; Aursand, Standal, Praël, McEvoy, & Axelson, 2009; Hidalgo & Zamora, 2003). Triacylglycerol (TAG) fatty acid composition in muscle of fatty fish, such as salmon, and in livers of gadoid fish, such as cod, reflects the diet (dos Santos, Burkow, & Jobling, 1993; Lie, 1991), which makes it possible to discriminate between wild and farmed fish from lipid analysis (Aursand, Mabon, & Martin, 2000; Aursand et al., 2009; Standal, Praël, McEvoy, Axelson, & Aursand 2008). However, fatty acid composition of fish is a net result of a wide range of factors, including, lipid metabolism, season, age, size and stage of sexual maturity and environmental factors (Sargent, Bell, McEvoy, Tocher, & Estevez, 1999). It has also been shown that certain fatty acids seem to be maintained within limits which depend on the species (Aursand, Jørgensen, & Grasdalen, 1995). Mus-
609
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
cle of the lean marine fish species consists primarily of phospholipids (PLs), and even though polar lipids are affected by diet to some degree (Mørkøre, Netteberg, Johnson, & Pickova, 2007), they are less influenced by diet than TAGs (Lie, Hemre, & Lambertsen 1992; Nanton, Lall, & McNiven, 2001). Therefore, when studying species or stock differences, PL analysis is recommended (Joensen, Steingrund, Fjallstein, & Grahl-Nielsen, 2000). Analyses on eggs from various cod stocks, suggested that fatty acid compositions of PLs were more dependent on stocks then on diet (Pickova, Dutta, Larsson, & Kiessling, 1997). Species of Atlantic tuna have been distinguished from PL–fatty acid profiles of white muscle (Medina, Aubourg, & Pérez Martín, 1997). The standard analysis of fatty acids is by gas chromatography (GC), but the use of high resolution (HR) 13C nuclear magnetic resonance (NMR) spectroscopy in the analysis of lipids is increasing. 13 C NMR is a multicomponent technique, which gives, in addition to fatty acid composition as GC, information on lipid classes and sn-2 position specificity of fatty acids in TAGs, which are characteristic for certain species (Aursand et al., 1995), and PLs (Falch, Størset, & Aursand, 2006). Previous analyses of marine oils showed that one could easily distinguish oils of different species according to their TAG profile obtained by 13C NMR (Aursand et al., 1995; Standal, Axelson, & Aursand, 2009). 13 C NMR studies on marine PLs include analysis on cod gonads (Falch, Størseth, & Aursand, 2007; Falch et al., 2006), and tuna muscle PLs (Medina & Sacchi, 1994; Sacchi et al., 1993). However, to the authors’ knowledge, 13C NMR studies of muscle lipids extracted from different species and stocks of lean gadoid fish, have not been published previously. The aim of this study was to evaluate if 13C NMR phospholipid profiles are characteristic enough to discriminate fish species and stocks. For the study five categories of lean gadoid fish were selected, namely: two stocks of Atlantic cod (Gadus morhua L.) (north-east arctic cod and Norwegian coastal cod), and the species haddock (Melanogrammus aeglifinus), saithe (Pollachius virens), and pollack (P. pollachius). The 13C NMR spectra were investigated to observe any differences in stereospecific distribution of the fatty acid 22:6n-3 in phosphatidyl choline (PC) and phosphatidyl ethanolamine (PE) among the different groups analysed. The full 13C NMR fingerprints were applied in multivariate analysis to observe groupings, to classify samples according to species/stocks, and to determine important variables in the classification. 2. Materials and methods 2.1. Materials The following species and stocks of lean fish were caught outside the coast of Vikna, Nord-Trondelag, Norway during March/ April 2006 (in the traditional fishing season): north-east arctic cod (AC, n = 6) and Norwegian coastal cod (CC, n = 6) (G. morhua L.), haddock (H, n = 6) (M. aeglefinus), saithe (S, n = 5) (P. virens), pollack (P, n = 4) (P. pollachius). The fish was frozen within 24 h by the fisherman and transported to the laboratory. In total, 27 fish were analysed. The average round weight of the different fish categories (with standard deviation) are north-east arctic cod: 2.9 ± 0.6 kg, coastal cod: 2.2 ± 0.3 kg, haddock: 0.7 ± 0.2 kg, saithe: 2.7 ± 1.2 kg and pollack: 4.3 ± 1.1 kg.
2.3.
13
C NMR
Approximately 90 mg of the oil sample was transferred to 5 mm NMR tubes and diluted with 0.6 mL deuteriated chloroform (CDCl3, 99.8% purity, Isotec Inc., Matheson). The 13C NMR spectra were run semi-quantitatively with a high number of scans, to achieve sufficient signal to noise (S/N) ratio to evaluate the stereospecific distribution of fatty acids in PLs. Comparisons between intensities of resonances within the same spectra should be performed with care in semi-quantitative 13C NMR spectra, however, comparison of signals among the different spectra is possible. Previous studies have shown that a semi-quantitative approach can be applied in positional distribution measurements, since the nuclear overhauser effect (NOE) effect is similar for carbonyl-carbons and that the T1 values of these carbons does not vary much according to the position of the fatty acids in the glycerol molecule (Aursand et al., 1995; Wollenberg, 1990). 13C NMR spectra were obtained on a Bruker Avance DMX 600 instrument (Bruker BioSpin GmbH, Rheinstetten, Germany) operating at 150.9 MHz for carbons, using a 5 mm BBO probe at 298 K. Power gated decoupling was applied. The following acquisition parameters were used: time domain 64k, pulse width 60°, sweep-width 200.8 ppm, acquisition time 1.08 s, relaxation delay 0.5 s, number of scans 16k. Zero filling and exponential line broadening (0.15 Hz) were applied before Fourier transform. The chemical shift scale is referred indirectly to TMS by the triplet of CDCl3 at 77.11 ppm. 2.4. Peak picking of full
13
C NMR spectra
Peak positions and intensities were obtained for resonances >0.5% of the maximum peak intensity within each spectrum. The resulting peak list was exported for manual alignment due to small variations in chemical shifts among samples. The solvent resonances were removed, before the data matrix (254 variables for the 27 samples investigated) were exported for multivariate analysis. 2.5. Calculation of the sn-2 position specificity of 22:6n-3 in PC and PE Peakfitting was applied to the carbonyl region of the 13C NMR spectra, to facilitate integration of peaks arising from 22:6n-3 in sn-1,3 and sn-2 position of PC and PE and to reveal hidden peaks (in particular 22:6n-3 in sn-2 position of PE consisted of more than one peak). The 13C NMR data were processed in xwinnmr as previously described, and the full spectra (1r file) were converted into ASCII files. Data in the carbonyl region (174–172 ppm) were selected (681 data points) and imported to PeakFit 4.12 (SeaSolve Software Inc, San Jose, CA). The AutoFit Residuals method was chosen as peak-fitting method. This procedure initially places peaks by finding local maxima in a smoothed data stream. Hidden peaks are than optionally added where peaks in the residuals occur (Peakfit 4.12). Peaks were fitted, assuming Lorentzian lineshape, with a linear baseline substracted prior to fitting. The area of the peaks arising from sn-1 and sn-2 22:6n-3 in PC and PE were registered, and the relative distribution of 22:6n-3 in sn-2 position calculated. Comparisons between means for the five fish categories were performed by one-way analysis of variance (ANOVA) with a significance level of 0.01 (Excel, Microsoft XP). 2.6. Multivariate analysis
2.2. Lipid extraction Lipid was extracted from white fish muscle under the back dorsal fin according the Bligh and Dyer method (1959). Before analysing the lipid extract by NMR, parts of the chloroform phase were removed by evaporation.
The 13C NMR data were analysed by the multivariate methods principal component analysis (PCA) (Jolliffe, 1986; Wold, Esbensen, & Geladi, 1987), linear discriminant analysis (LDA) and Bayesian belief network (BBN) (Heckerman, Geiger, & Chickering, 1995; Pearl, 1988) to observe groupings and to test if it is possible to
610
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
discriminate fish of different species (cod, haddock, saithe, pollack) and stocks (north-east arctic cod, Norwegian coastal cod). PCA was applied as an unsupervised multivariate technique. PCA has frequently been applied to spectral data for dimensionality reduction, to identify outliers and to classify samples. In PCA the original variables are transformed into new, uncorrelated variables called principal components, which retains as much as possible of the information present in the original data. Each principal component (PC) is a linear combination of the original variables. The scores of a subset of the principal components can be used in subsequent multivariate analysis. LDA is a supervised pattern recognition technique, which is based on the assumption that samples of the same group are more similar than samples belonging to different groups. The technique seeks to find a linear transformation by maximising the betweenclass variance and minimising the within-class variance. LDA has been widely used for pattern recognition and data analysis. In situations where the number of variables exceeds the number of ob-
AC1
jects, the principal components can be used as variables in the linear discriminant analysis (Kim, Kim, & Bang, 2003). BBN analyses have been used extensively in the area of medical diagnostics and bioinformatics, and recent studies have also shown the potential of this technique for analyses of HR NMR data (Martinez et al., 2009). A Bayesian network consists of: (1) nodes which represent the random variables, where each node has states (a set of probable values for each variable), (2) directed edges (arrows) which connect the nodes (represent dependencies, absence of arrows indicate independence), (3) a conditional probability table which is associated with each node (prior probability), and (4) a directed acyclic graph where the graph represents independence relationships between variables. The most important advantages for this work are that BBNs calculate explicit probabilities for a hypothesis, can be used as a classifier and can find the variables with the most impact on the classification. A more thorough description of the BBN method is given in (Axelson, Standal, Martinez, & Aursand, 2009).
PC (sn-3)
PC (CH2O)
PE/PC (sn-1)
PL (sn-2) TAG (sn-1,3)
PE (CH2O)
TAG (sn-2)
70
69
68
67
66
65
64
63
ppm
70
69
68
67
66
65
64
63
ppm
70
69
68
67
66
65
64
63
ppm
70
69
68
67
66
65
64
63
ppm
70
69
68
67
66
65
64
63
ppm
CC1
H1
S5
P4
Fig. 1. Glycerol region of 13C NMR spectra of total lipids extracted from white muscle of north-east arctic cod (AC), Norwegian coastal cod (CC), haddock (H), saithe (S) and pollack (P).
611
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
The NMR data were normalised to an area of 100, mean centred and autoscaled prior to multivariate analyses. PCA was carried out (Minitab15, Minitab Inc.) by using the 254 chemical shift intensities as input data (with chemical shifts from solvents excluded in analysis). In the LDA (Minitab, version 15), the sample scores on PC components 1–3 in the previously mentioned PCA calculations, were used as input variables. The analyses were performed with sample grouping 1–5 according to species/stock (north-east arctic cod, Norwegian coastal cod, haddock, saithe, pollack). Cross validation (CV) was used as validation method when LDA models were developed. In the leave one out (LOO) CV each sample is removed one at a time, the classification function is recalculated using the remaining data, and the omitted observation is classified. BBN predictions (Netica v4.02, Norsys Sofware Corp. Vancouver B.C., Canada) were made for 18 samples that were randomly se-
lected in each group for the training set, while nine were held back for the validation calculations. The variables with most impact on the classification were identified from the BBN analysis, and corresponding NMR peaks were assigned according to previous analysis on marine PLs (Falch et al., 2006, 2007).
3. Results and discussion 3.1.
173.6
173.4
22:6n-3 (sn-2) PC
20:5n-3 and 20:4n-6 (sn-2) PC
MUFA/SFA (sn-1,3) MUFA/SFA (sn-2) TAG PL
173.8
C NMR spectra
Assignments of the 13C NMR spectra of total lipids extracted from fish muscle in this study were made according to previous studies on marine PLs (Falch et al., 2006, 2007). The glycerol and polar-headgroup region (61–71 ppm, Fig. 1) gives information
MUFA/SFA (sn-1) PL
AC1
13
173.2
22:6n-3 (sn-1) 22:6n-3 PC (sn-1) PE
173.0
22:6n-3 (sn-2) PE
172.8
172.6
172.4
172.2
ppm
MUFA/SFA (sn-2) TAG and 20:5n-3 (sn-2) PE
CC1
173.8
173.6
173.4
173.2
173.0
172.8
172.6
172.4
172.2
ppm
173.8
173.6
173.4
173.2
173.0
172.8
172.6
172.4
172.2
ppm
173.8
173.6
173.4
173.2
173.0
172.8
172.6
172.4
172.2
ppm
173.8
173.6
173.4
173.2
173.0
172.8
172.6
172.4
172.2
ppm
H1
S5
P4
Fig. 2. Carbonyl region of 13C NMR spectra of total lipids extracted from white muscle of north-east arctic cod (AC), Norwegian coastal cod (CC), haddock (H), saithe (S) and pollack (P).
612
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
from cholesterol could be found in other parts of the spectra (Falch et al., 2006). Lipids from white muscle of lean gadoid fish contain approximately about 73–78% PLs, 3–9% TAGs, and 6–9% sterols (dos Santos et al., 1993). Even though the glycerol- and polar-headgroup region (Fig. 1) shows characteristics according to stereospecific conformation, the carbonyl region is better suited for the purpose of obtaining data on the stereospecific distribution of fatty acids (Fig. 2). From Fig. 2, one can see variation in relative peak intensities among the five fish categories. Most pronounced is the variation in the peak intensity arising from sn-2 20:5n-3/20:4n-6 in PC at 172.95 ppm. However, this variation in sn-2 20:5n-3/20:4n-6 content is presumably not species-specific, as large intra-specific differences of the same fish species (e.g. 5 coastal cods) was observed (results not shown). The fish in this study were caught in the traditional fishing season of these species, however this is identical with the spawning time, and the intra-specific differences in lipid composition may be due to variations in reproductive status, but also life stage, size, sex, food availability and environmental factors may influence fatty acid composition (Sargent et al., 1999).
Table 1 Average values on the sn-2 position specificity of 22:6n-3 in PC and PE (mole%). AC: north-east arctic cod, CC: Norwegian coastal cod, H: haddock, S: saithe, P: Pollack.
AC CC H S P
n=6 n=6 n=6 n=5 n=4
DHA sn-2 PC
DHA sn-2 PE
84 ± 3 78 ± 3 85 ± 3 80 ± 3 77 ± 3
76 ± 3 71 ± 1 71 ± 5 76 ± 5 65 ± 4
about lipid classes present in the sample. Signals from the principal PL species in cod muscle, PC and PE, but also traces of TAGs could be seen in the spectra (Fig. 1). Even though there were minor differences in the level of triacylglycerols (TAGs) among the samples (as one can also observe for the five samples in Fig. 1), the level of triacylglycerols was considered too low to have a significant impact on the fatty acid composition. As an example; only one sample (of saithe) showed relatively high levels of TAGs (approximately twice the level of sample S5 in Fig. 1), but still displayed similar fatty acid composition as the other saithes. Several resonances
(a)
AC3
10 AC1 AC5 AC2 AC6 AC4
PC2
5
0
CC5 S1 CC6 CC2 S2 CC4 S4
S5
H4
H5 CC1
P2
-5
H6
H2 CC3
H1
H3
P3 P4
P1 S3
-10 -10
(b)
0
10 PC1
20
30
10 CC4 CC2
5
H4
CC5 P2 P3
CC6 H3
P1
AC6 AC2 AC4AC3 AC5 AC1
S5
H5
H2 CC1
PC3
0
CC3
P4
H1
H6 S4
-5
S3 S2
-10 S1
-15 -10
0
10 PC1
20
30
Fig. 3. PC1 vs. PC2 (a) and PC1 vs. PC3 (b) score plot from NMR data for the five different groups of fish. The three PCs explain 36%, 9% and 9% of the variance in the dataset respectively. AC: north-east arctic cod, CC: Norwegian coastal cod, H: haddock, S: saithe, P: pollack (254 chemical shifts as input variables).
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
3.2. Stereospecific calculations As previously mentioned, the stereospecific distributions of TAGs in fish lipids are quite specific for certain species (Aursand et al., 1995). To evaluate if similar differences could be found in marine PLs, the sn-2 position specificity of 22:6n-3 in PC and PE was calculated from the 13C NMR spectra. Well resolved peaks from 22:6n-3 in sn-2 and sn-1 position in PE and PC appeared in the carbonyl region of the 13C NMR spectra (Fig. 2). This allowed the calculation of the stereospecific distribution of this fatty acid in PC and PE. Average values on the sn-2 position specificity of 22:6n-3 for each species/stocks are given in Table 1. The results showed that 22:6n-3 is mainly esterified to sn-2 of PC in all species, with average values between 77% and 85% for the five categories. Also in PE the sn-2 position was the preferred position for 22:6n-3, with average values for the five groups ranging from 65–76%. These results are in accordance with previous studies on PLs in cod milt and roe, which showed that the n3 polyunsaturated fatty acids (PUFAs) were primarly esterified in the sn-2 position of PC and PE (Bell & Dick, 1991; Falch et al., 2006). In TAGs of cod liver oil, DHA was concentrated in sn-2 posi-
613
tion with 74.4%, while EPA was nearly randomly distributed with 39.7% (Aursand et al., 1995). The pollack group displayed the lowest values of 22:6n-3 in sn2 position both for PE and PC. ANOVA calculations showed that the ratio of DHA in sn-2 PC was significantly lower for the pollacks than for the north-east arctic cods and the haddocks. Also the coastal cod group displayed lower mean value than the haddocks for DHA in sn-2 PC. For PE, the ratio of sn-2 22:6n-3 of was significantly lower for the pollack group than for north-east arctic cods, coastal cods and the saithes. Further studies are needed to conclude whether these differences, are due to genetic differences, or if stereospecific distribution of fatty acids is influenced by factors such as diet, sex and reproductive status. The relatively low values on 22:6n-3 in sn-2 position of PE and PC in pollacks may also be attributed to the generally higher level of 22:6n-3 in these fish (and thereby possibly higher level of 22:6n-3 in sn-1 position) compared to the other fish in this study. A weak linear correlation (negative) were found between DHA levels and DHA in sn-2 positions of PC ( 0.43) and PE ( 0.34). However, to which extent dietary fatty acid composition influence the stereospecific distribution of fatty acids remains to be investigated.
Fig. 4. Distributions of the eight most important variables in the BBN analysis for the classes AC (top) and CC (bottom). Horizontal bars denote relative percentage of values of the given variable (chemical shift, ppm).Values at the bottom are the mean and standard deviation for the chemical shift intensities. AC: north-east arctic cod, CC: Norwegian coastal cod, H: haddock, S: saithe, P: pollack.
614
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
29.29; 29.31; 29.32
34.5
34.0 128.6
28.04
28.14
33.96
29.11
ppm
29.0
28.5
ppm
128.02 128.27 70.77 127.54
128.5 Fig. 5. The
128.0
ppm
71.5
71.0
70.5
ppm
13
C NMR spectrum of north-east arctic cod (CC) with some of the most important variables in the BBN classification highlighted.
3.3. Multivariate analysis on
13
C NMR data
PCA was applied to the NMR data to observe groupings and to interpret differences among groups. The PCA was carried out with 254 chemical shifts as input variables. The score plot of PC1 vs. PC2 is shown in Fig. 3a. Samples with similar PL profiles will be closely located in this score plot. The first two principal components account for 36% and 9% of the variance in the dataset. Clustering of fish of the different groups can be observed; the north-east arctic cod (AC) samples are well separated from the other samples in this score plot, and the pollack (P) samples are also grouped together. However, the coastal cod (CC) samples are positioned far away from each other, and overlaps with the saithes (S) and the haddocks (H). However, also PC3 (explaining 9% of the variance) contains valuable information when it comes to separating the five different classes from NMR data. In this PCA plot, also the saithe (S) samples are closely positioned (Fig. 3b). North-east arctic cod migrates from the feeding areas in the Barents Sea to the spawning areas along the Norwegian coast in March/April, in contrast to the coastal cod, which are quite stationary. Coastal cod is listed as an endangered species on the ‘‘red list” of Norwegian Biodiversity Information Centre (www.artsdatabanken.no). The fact that the north-east arctic cod samples are completely separated from the Norwegian coastal cods in Fig. 3a, indicates that there are differences in lipid composition between these two stocks of Atlantic cod, which may arise from dietary or metabolic differences. The two stocks have previously been distinguished from otholite-, haemoglobin-, and gene analysis (Sarvas & Fevolden, 2005). For a more quantitative result on the classification power of the dataset, LDA and BBN predictions were applied. The three first scores from the previous PCA were used as input variables in the LDA. LDA with the 3PCs obtained from NMR data gave 21/27 correct classifications (78% correctly classified). The BBN approach also allows one to rank the variables according to their relevance to the classification. In the BBN analysis, 100% correct classifications were achieved with the 17 most important chemical shifts as variables. The 17 variables with the most impact on the classification were determined to be: 172.81, 26.56, 33.96, 29.31, 25.61, 29.29, 128.02, 29.14, 128.60, 70.77, 29.32, 28.04, 128.27, 29.11, 25.32, 127.54 ppm. In a subsequent reduction of variables, only one sample was misclassified with eight chemical shifts as variables (one of the coastal cod samples classified as north-east arctic cod in the validation dataset).
A schematic representation of BBN for the classification is given in Fig. 4 (example north-east arctic cod (AC) and Norwegian coastal cod (CC) groups). Horizontal bars denote relative percentage of values of the eight variables with the highest impact in the classification for samples of the indicated class (in this example: AC and CC). Values at the bottom are the mean and standard deviation for the given chemical shift. Assignment of important chemical shifts in the classification of the five fish categories were made according to Falch et al., 2006. The peak at 172.81 ppm (Fig. 2) arise from carbonyl-carbon of 22:6n-3 in sn-1 PC. Peaks around 33.96 ppm stem from 22:6n-3 sn-2 PL, while the singlet at 28.04 ppm arise from carbon number 19 of cholesterol (Fig. 5). However, other peaks were more difficult to identify. In example, the peaks at 29.32, 29.31, and 29.29 ppm (Fig. 5) arise from carbon atoms of various saturated, mono-and di unsaturated-fatty acids.
3.4. Conclusion This first screening indicates that it is possible to discriminate amongst species and stocks of lean gadoid fish based on lipid profiling of muscle lipids, which are mainly composed of phospholipids. Lipid profiles from 13C NMR gave 100% classification of the five categories of fish by the BBN approach. In a further evaluation of the reliability of this approach for authentication purposes, it would be important to map the intra-specific variation in the phospholipid profiles due to factors such as diet, year of catch, season, reproductive status, sex and geographical origin. If the profiles are characteristic of species regardless of such factors, it would, as for most authentication methods, be necessary to establish an extensive database with reference material covering this natural variation. 13 C NMR is a less sensitive technique than GC, but has the advantage that it can be applied non-destructively on intact lipids. Additional information is obtained from 13C NMR than from GC, such as information on lipid classes and the stereospecific distribution of fatty acids, which might be relevant for authentication purposes. Results on sn-2 position specificity of 22:6n-3 in PC and PE did show that there were significant differences between groups of fish, however, further studies are needed to conclude whether these differences can be attributed as genetic differences, and to which extent the stereospecific distribution of fatty acids is influenced by factors such as diet, sex and reproductive status.
I.B. Standal et al. / Food Chemistry 121 (2010) 608–615
Acknowledgements The Norwegian Research Council is gratefully acknowledged by the financial support to NFR-project no. 146932/130. Thanks also to Terje Breivik for collection of the fish. References Aursand, M., Jørgensen, L., & Grasdalen, H. (1995). Positional distribution of x3 fatty acids in marine lipid triacylglycerols by high-resolution 13C nuclear magnetic resonance spectroscopy. Journal of the American Oil Chemists’ Society, 72, 293–297. Aursand, M., Mabon, F., & Martin, G. J. (2000). Characterization of farmed and wild salmon (Salmo salar) by a combined use of compositional and isotopic analysis. Journal of the American Oil Chemists’ Society, 77, 659–666. Aursand, M., Standal, I. B., & Axelson, D. E. (2007). High-resolution 13C NMR spectroscopy pattern recognition of fish oil capsules. Journal of Agricultural and Food Chemistry, 55, 38–47. Aursand, M., Standal, I. B., Praël, A., McEvoy, L., & Axelson, D. E. (2009). 13C NMR pattern recognition techniques for the classification of Atlantic salmon (Salmo salar L.) according to their wild, farmed and geographical origin. Journal of Agricultural and Food Chemistry, 57, 3444–3451. Axelson, D. E., Standal, I. B., Martinez, I., & Aursand, M. (2009). Classification of wild and farmed salmon using Bayesian belief networks and gas chromatography derived fatty acid distributions. Journal of Agricultural and Food Chemistry, 57, 7634–7639. Bell, M. V., & Dick, J. R. (1991). Molecular-species composition of the major diacyl glycerophospholipids from muscle, liver, retina and brain of cod (Gadus morhua). Lipids, 26, 565–573. Bligh, E. G., & Dyer, W. J. (1959). A rapid method of total lipid extraction and purification. Canadian Journal of Biochemistry and Physiology, 37, 911–917. dos Santos, J., Burkow, I. C., & Jobling, M. (1993). Patterns of growth and lipid deposition in cod (Gadus morhua L.) fed natural prey and fish-based feeds. Aquaculture, 110, 173–189. EC Commission Regulation No. 2065/2001 of October 2001 laying down detailed rules for the application of Council Regulation (EC) No. 104/2000 as regards informing consumers about fishery and aquaculture products. Falch, E., Størset, T., & Aursand, M. (2006). Multi-component analysis of marine lipids in fish gonads with emphasis on marine phospholipids using high resolution NMR spectroscopy. Chemistry and Physics of Lipids, 144, 4–16. Falch, E., Størseth, T. R., & Aursand, M. (2007). High resolution NMR for studying lipid deterioration in cod (Gadus morhua) gonads. Chemistry and Physics of Lipids, 147, 46–57. Fiskeridirektoratet (2007).
615
Jolliffe, I. T. (1986). Principal component analysis. Berlin: Springer-Verlag. Kim, H. C., Kim, D., & Bang, S. Y. (2003). Extensions of LDA by PCA mixture model and class-wise features. Pattern Recognition, 36, 1095–1105. Lie, Ø. (1991). Studies on digestion, deposition and fatty acid composition of lipids in cod (Gadus morhua). PhD thesis, University of Bergen, Bergen, Norway. Lie, Ø., Hemre, G. I., & Lambertsen, G. (1992). Influence of dietary fatty acids on the glycerophospholipid composition in organs of cod (Gadus morhua). Lipids, 27, 770–775. Martinez, I., Aursand, M., Erikson, U., Singstad, T. E., Veliuylin, E., & Van der Zwaag, C. (2003). Destructive and non-destructive analytical techniques for authentication and composition analyses of foodstuffs. Trends in Food Science and Technology, 14, 489–498. Martinez, I., Standal, I. B., Axelson, D. E., Finstad, B., & Aursand, M. (2009). Identification of the farm origin of salmon by fatty acid and HR 13C NMR profiling. Food Chemistry, 116, 766–773. Medina, I., Aubourg, P., & Pérez Martín, R. (1997). Species differentiation by multivariate analysis of phospholipids from canned Atlantic tuna. Journal of Agricultural and Food Chemistry, 45, 1997–2499. Medina, I., & Sacchi, R. (1994). Acyl stereospecific analysis of tuna phospholipids via high resolution 13C NMR spectroscopy. Chemistry and Physics of Lipids, 70, 53–61. Mørkøre, T., Netteberg, C., Johnson, L., & Pickova, J. (2007). Impact of dietary oil source on product quality of farmed Atlantic cod, Gadus morhua. Aquaculture, 267, 236–247. Nanton, D. A., Lall, S. P., & McNiven, M. A. (2001). Effects of dietary lipid level on liver and muscle lipid deposition in juvenile haddock, Melanogrammus aeglefinus L. Aquaculture Research, 32, 225–234. Norwegian Biodiversity Information Centre.