1
Investigation of production method, geographical origin and species authentication in commercially
2
relevant shrimps using stable isotope ratio and/or multi-element analyses combined with chemometrics:
3
an exploratory analysis
4
5
Ignacio Orteaa,b*
6
José M. Gallardoa
7
8
a
9
Vigo, Spain.
Department of Food Technology, Marine Research Institute (IIM), Spanish National Research Council (CSIC),
10
b
11
2440 Geel, Belgium.
12
*Corresponding author: nachoog@iim.csic.es; telephone +32 14571829; fax+32 14571787
Present address: Institute for Reference Materials and Measurements (EC-JRC-IRMM), Retieseweg 111, B-
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
1
31
Abstract
32
33
Three factors defining the traceability of a food product are production method (wild or farmed),
34
geographical origin and biological species, which have to be checked and guaranteed, not only in order to avoid
35
mislabelling and commercial fraud, but also to address food safety issues and to comply with legal regulations.
36
The aim of this study was to determine whether these three factors could be differentiated in shrimps using
37
stable isotope ratio analysis of carbon and nitrogen and/or multi-element composition. Different multivariate
38
statistics methods were applied to different data subsets in order to evaluate their performance in terms of
39
classification or predictive ability. Although the success rates varied depending on the dataset used, the
40
combination of both techniques allowed the correct classification of 100% of the samples according to their
41
actual origin and method of production, and 93.5% according to biological species. Even though further studies
42
including a larger number of samples in each group are needed in order to validate these findings, we can
43
conclude that these methodologies should be considered for studies regarding seafood product authenticity.
44
45
Keywords
46
Food authenticity; geographical origin; production method; seafood authentication; species authentication
47
48
Abbreviated running title
49
Shrimp authentication by stable isotope ratio and element analyses.
50
51
52
53
54
55
56
57
58
59
60
2
61
1. Introduction
62
63
Fishery products, especially Decapoda crustaceans, are among the most relevant food commodities in
64
terms of trading figures. Among them, marine shrimps and prawns account for about 15% of the total value of
65
internationally traded fishery products (FAO Fisheries and Aquaculture Department, 2012) and are important
66
economic resources for many countries. In addition to catches in different areas—mainly the Food and
67
Agriculture Organization of the United Nations (FAO) fishing areas 51 (Western Indian ocean) and 71 (Western
68
Central Pacific ocean)—shrimps and prawns are increasingly being produced in aquaculture facilities, especially
69
in some areas of China, Southeast Asia and Ecuador.
70
Due to the increasing demand for foodstuffs in general, and for high-quality products in particular, and
71
also due to the globalisation of markets and trade, adulteration can occur through the food chain (Pascoal,
72
Barros-Velázquez, Cepeda, Gallardo & Calo-Mata, 2008a). A product can be deliberately substituted with a
73
lower quality and cheaper counterpart, or an unintentional error can cause an inadvertent mislabelling of
74
products, leading, in both cases, to commercial fraud that affects both the food industry and the consumers.
75
Adulteration of food products not only has economic implications, but also represents a potential public health
76
risk (Spink & Moyer, 2011). Regulations have appeared all over the world in order to fight against adulteration
77
and misbranding of foods (U.S. Food and Drug Administration, 2013), underlining the need for labelling fishery
78
and aquaculture products with the scientific name of the species used, the production method (caught at sea or
79
from aquaculture), and the place of origin (Council Regulation (EC) No 104/2000 of 17 December, 1999).
80
Among fish products in general, and crustacean species in particular, the substitution of an appreciated
81
high quality species by another of lower quality and with a lower price is especially frequent (Pascoal et al.,
82
2008a). In these cases, inadvertent or deliberate adulteration leads to mislabelling and commercial fraud. In
83
addition, these adulteration practices can affect the marine conservation programs that protect overexploited
84
species or populations (Rasmussen & Morrisey, 2008).
85
The place of origin of the components within a food product should also be checked and guaranteed.
86
Consumers demand information about the geographic origin of the food they eat. In addition to the perception of
87
higher quality products with geographical indications (Protected Designation of Origin, Protected Geographical
88
Indication, Traditional Specialty Guaranteed), some other geographic-related factors affect consumer habits, and
89
therefore the food industry, trade and regulations. These factors include the consumer’s preference for
90
foodstuffs they perceive as having a lower environmental footprint (e.g., local products), the perception of
3
91
different health safety implications depending on the origin of the product, and even patriotism in the sense that
92
some people feel they should buy food produced in their own country. Origin authenticity assessment is
93
important not only because of the increasing demand for information from consumers and to prevent
94
commercial fraud, but also, in order to ensure the safety of foodstuffs, traceability at all stages is compulsory for
95
all food and feed industries (Regulation (EC) No 178/2002 of the European Parliament and of the Council of 28
96
January, 2002). The relevance of traceability for food quality and safety has been clearly demonstrated after
97
recent serious food crises, such as the 2013 scandal of non-beef that was labelled beef mincemeat in Europe
98
(Quinn, 2013); the 2011 E. coli outbreak in Germany (Buchholz et al., 2011); the recurring bird flu events in
99
China (Larson, 2013) and the epidemic of bovine spongiform encephalopathy in the UK (Will et al., 1996). It
100
must be pointed out that seafood in particular is exposed to a wide spectrum of contaminants, such as pathogenic
101
microorganisms, heavy metals and chemical pollutants, and therefore the authentication of its geographical
102
origin is of capital importance, especially when a contaminated product from a particular area must be
103
withdrawn from the market.
104
Together with the species and the geographical origin, another important factor defining traceability is
105
the production method (e.g., wild or farmed; organic or intensive) (Moretti, Turchini, Bellagamba & Caprino,
106
2003). During recent years, farmed fish production has greatly increased compared to wild fish capture.
107
However, organoleptic characteristics, nutritional values and usually price are not the same for fish that is wild-
108
caught and fish that is farmed. Furthermore, the quality of cultured fish depends on the diet they receive.
109
For all the reasons stated above, the authentication and origin of foodstuffs must be guaranteed, and accurate
110
and reliable analytical methods are needed in order to verify the production method, geographical origin and
111
species of the components used in a food product. Many different instrumental techniques have been proposed
112
for food authentication (Drivelos & Georgiou, 2012): high-performance liquid chromatography (HPLC), nuclear
113
magnetic resonance (NMR) spectroscopy, infrared spectroscopy (IR), capillary electrophoresis, and, more
114
recently, DNA-based and proteomic methods (Ortea, Pascoal, Cañas, Gallardo, Barros-Velázquez & Calo-Mata,
115
2012). Over the last decade, the isotope ratio and the composition of selected elements have provided
116
measurements revealing a unique isotopic fingerprint for the sample being analysed, so stable isotope ratio and
117
multi-element analysis have become two of the most frequently used techniques for assessing authenticity and
118
traceability in food products (Drivelos et al., 2012).
119
120
The content and availability of elements in soils depends on different factors, such as pH, humidity and
clay-humus complex (Kim & Thornton, 1993). The elemental composition of vegetation reflects the bio-
4
121
available and mobilised nutrients present in the soils where plants are growing, which also determines the multi-
122
element composition of the animals consuming that vegetation. Therefore, the elemental composition profile
123
may be used as a unique marker for characterising the diet and geographical origin of food products (Kelly,
124
Heaton & Hoogewerff, 2005). Due to the differential distribution of C3 and C4 plants from the equator to the
125
poles, and due to the fact that C3 plants have lower 13C/12C ratios than C4 plants, differences in 13C/12C ratios can
126
be found in plant material according to geographical origin (Kelly et al., 2005). Differences in 15N/14N ratios are
127
more related to local agricultural practices (Oulhote, Le Bot, Deguen & Glorennec, 2011). Since carbon and
128
nitrogen propagate from prey to predator through the trophic chain, 13C/12C and 15N/14N isotope ratios yield a
129
history of feeding relationships. They are the two most informative parameters for analysing the diet of animals,
130
and, therefore, they can be used as a proxy for determining geographical origin.
131
Stable isotope ratio (SIR) and multi-element analyses have been used extensively for authenticating
132
wines, fruit juices, olive oil, honey, milk and dairy products, coffee, tea, cereal crops and meat (Drivelos et al.,
133
2012; Gonzálvez, Armenta & de la Guardia, 2009; Kelly et al. 2005; Vinci, Preti, Tieri & Vieri, 2013), but have
134
scarcely been used in fish and shellfish studies. In seafood species, most of these studies have applied SIR
135
analysis to discriminate economically important wild and farmed fish species, such as Atlantic salmon
136
(Dempson & Power, 2004), gilthead sea bream (Moreno-Rojas, Serra, Giani, Moretti, Reniero & Guillo, 2007;
137
Serrano, Blanes & Orero, 2007), turbot (Busetto et al., 2008) and sea bass (Gordon Bell et al., 2007), and even
138
to differentiate farmed trout fed with plant- or fish-protein-based diets (Moreno-Rojas, Tulli, Messina, Tibaldi &
139
Guillou, 2008). Few reports are available regarding the use of element profiling to establish the geographical
140
origin of marine organisms. Smith and Watts (2009) determined the origin of farm-raised shrimp by trace metal
141
profiling and multivariate statistics using a database on the composition of shrimp from different countries,
142
including white leg shrimp (Penaeus vannamei) and black tiger shrimp (Penaeus monodon). Wild and farm-
143
raised salmon were differentiated using elemental analysis and different classification modelling (Anderson,
144
Hobbie & Smith, 2010), and sea cucumber samples from three water environments of China were identified
145
using multi-element analysis and chemometric techniques (Liu, Xue, Wang, Li, Xue & Xu, 2012). Regarding
146
species identification, the SIRs of carbon and nitrogen were recently used to differentiate the gadoid fish species
147
Atlantic cod and saithe (Monteiro Oliveira, Sant'Ana, Ducatti, Denadai & de Souza Kruliski, 2011). Due to the
148
large number of samples and variables usually measured in these profiling strategies, the application of the
149
appropriate chemometric tools is essential to extract relevant information and establish patterns between
5
150
different samples. Multivariate data analysis methods are highlighted as the most frequently used, especially
151
principal component analysis (PCA) and discriminant analysis (Drivelos et al., 2012; Oulhote et al., 2011).
152
The aim of this study was to determine whether shrimp species, geographical origin and production
153
method (wild vs. farmed) could be differentiated using either SIR analysis of carbon and nitrogen or multi-
154
element composition (Pb, Cd, As, P, S) or a combination of both. Results obtained using different datasets for
155
the element composition analysis (only Pb, Cd and As; only P and S; and the five elements together) were
156
compared. To assess for the effect of different sample sizes when a relatively low number of samples are
157
analysed by SIR, as it is done in many authenticity studies found in bibliography, we added a new set of
158
specimens to the initial set and compared the results obtained (n=31 vs n=45).
159
160
At each of the three levels (species, origin and production method), different multivariate statistics
methods were applied in order to evaluate their performance.
161
162
2. Materials and methods
163
164
2.1. Shrimp samples
165
166
Specimens analysed according to production method, geographical origin and species are shown in
167
Table 1. They were collected using either extractive fishing procedures or from aquaculture farms from different
168
geographical localisations worldwide. Shrimps, whole animals, were frozen on board and shipped to our
169
laboratory for the analyses. Specimens were classified in their respective taxons with the help of a marine
170
biologist from the Marine Sciences Institute (Mediterranean Centre for Marine and Environmental Research,
171
Spanish National Research Council, CMIMA-CSIC, Barcelona, Spain) with expertise in penaeid shrimp
172
taxonomy. In total, 45 shrimp individuals from different species, geographical origins and production methods
173
were analysed. Seven different shrimp species from the order Decapoda were considered in this study; five of
174
them were from the Penaeidae family, Penaeus monodon, Litopenaeus vannamei, Fenneropenaeus indicus,
175
Fenneropenaeus merguiensis, Farfantepenaeus notialis; one was from the Solenoceridae family, Pleoticus
176
muelleri, and one was from the Pandalidae family, Pandalus borealis. The samples were obtained from a total
177
of nine different geographical origins. Three of the species, namely F. notialis, P. monodon and L. vannamei,
178
were collected from different areas (F. notialis from two different origins, Senegal and Nigeria; P. monodon
179
from two different origins, Mozambique and farm b; and L. vannamei from farms a and c). The geographic
6
180
origin and species of every specimen analysed is shown in Table S1 in the Supplementary Material. Samples
181
were stored frozen in polyethylene bags.
182
183
2.2. Reagents
184
185
Chloroform and methanol were purchased from Panreac (Barcelona, Spain); nitric acid and hydrogen
186
peroxide were purchased from Merck (Darmstadt, Germany); and the reference material, TORT-2, was obtained
187
from the National Research Council of Canada.
188
189
2.3. SIR and multi-element analysis
190
191
White muscle samples were dried at 60 ºC for 72 h and were pulverised to a fine powder using a ball
192
mill grinder and stored in glass desiccation vials. For SIR analysis of 15N/14N and 13C/12C, 0.2 g of dried, ground
193
tissue from the 45 samples considered were defatted as described by Folch, Lees & Stanley (1957), with little
194
modifications. Briefly, 10 ml CHCl3/CH3OH 2:1 (v/v) were added, and after 10 min agitation, the solution was
195
kept at 4 ºC for 12 h. Samples were then centrifuged (750 rpm, 10 min) and supernatants were eliminated. After
196
three extractions, samples were dried for 12 h. Dried residue was cleared three times with 10 ml distilled water
197
and centrifuged for 10 min at 1200 rpm to eliminate the supernatant. The final pellet was dried at 50 ºC for 12 h
198
and pulverised again. Then, 1 to 2 mg of the homogeneous dried material was analysed using an isotope ratio
199
mass spectrometer (Thermo-Finnigan MAT 253, Bremen, Germany). All results were reported with respect to
200
the international reference standards PDB (PeeDee Belemnite) for δ 13C and atmospheric N2 for δ15N, according
201
to the equation:
202
δref = (Rsample – Rref / Rref) 1000,
203
204
205
where δref is the isotope ratio of the sample expressed in delta units (per mil) relative to the reference material,
206
and Rsample and Rref are the isotope ratios of the sample (e.g., 15N/14N or 13C/12C) and the reference material,
207
respectively. The standard deviations observed between replicates ranged from 0.03‰ for δ13C to 0.14‰ for
208
δ15N.
7
209
For element analysis, 0.25 g of each of the 31 samples considered were digested with 4 ml of
210
concentrated ultrapure HNO3 and 4 ml of H2O2 (50% v/v) in a microwave oven (Multiwave 3000, Anton Paar,
211
Graz, Austria). Digested samples were diluted to a final volume of 50 ml with double de-ionised water. Samples
212
were analysed with inductively coupled plasma-optical emission spectrometry (ICP-OES) (model Optima 4300
213
DV, Perkin-Elmer, Massachusetts, USA) for P and S, and an inductively coupled plasma-mass spectrometer
214
(ICP-MS) (model X7, Thermo Fisher Scientific, Waltham, USA) for As, Cd and Pb. Analytical quality was
215
assessed using a standard reference material from lobster hepatopancreas, TORT-2, which was treated and
216
analysed under the same conditions as the samples. Recoveries of all the elements ranged from 97% to 106% of
217
the certified value. Detection limits were determined as the concentration corresponding to three times the
218
standard deviation of ten blanks.
219
220
2.4. Statistical analysis
221
222
The statistical analysis of data was performed using the statistical package R version 3.0.1 (www.r-
223
project.org). At the three different levels (species, origin and production method) and using different datasets
224
(only SIR data, only multi-element composition, or a combination of both), different multivariate statistics
225
techniques were applied in order to visualize the data structure and to classify the samples, namely PCA, cluster
226
analysis, k-means classification and discriminant analysis (DA). In PCA, the different samples are grouped as a
227
function of different variables, called principal components, demonstrating differences among the studied
228
samples and reducing the complexity of the system. In cluster analysis, samples are grouped in clusters
229
according to their similarity or distance, assuming that the nearness of samples reflects the similarity of their
230
features. In k-means clustering, the aim is to determine the best possible partitioning of the samples into a
231
specific number of clusters that should be as homogeneous as possible. Finally, DA is used to evaluate whether
232
the different groups could be mathematically distinguished according to the different datasets. The proportion of
233
observations correctly allocated to the current group was evaluated by both empirical and cross-validation
234
methods. The ‘leave-one-out’ procedure was used for cross-validation to assure that the discriminant function is
235
acceptable for the population and not only for the analysed sample. In this procedure, the discriminant analysis
236
is repeated n times excluding one specimen each time, and this specimen is assigned using the discriminant
237
function obtained. In this way, all elements are assigned without being used in building the discriminant
238
function, and therefore all the sample is used as a validation sample. If the success rate obtained in the cross-
8
239
validation is similar to that obtained in the discriminant analysis of the entire sample, validation is positive and
240
the results are acceptable for the whole population.
241
242
3. Results and Discussion
243
244
245
246
The samples considered were subjected to SIR analysis of carbon and nitrogen, and to multi-element
composition (Pb, Cd, As, P, S) analysis. The data obtained are summarised in Table 1.
Data were represented in bivariate scatter plots (δ13C vs. δ15N; As vs. Cd; As vs. Pb; Pb vs. Cd and P
247
vs. S) (Fig. S1 in the supplementary material). From observing these bi-plots, we can conclude that representing
248
δ13C vs. δ15N (Fig. 1a) can help in the differentiation of those samples coming from aquaculture facilities, since
249
they are outside of the central group of wild samples. Further, one could initially classify the six samples of
250
Argentinian origin as ‘farmed’, since they are also located outside of the central group of ‘wild’ samples in the
251
plot (Fig. 1b). However, it has to be noted that it is the δ13C axis that provides the correct classification, and all
252
the samples in the range of δ13C -20 to -14, which includes the Argentinian samples, can be classified as ‘wild’.
253
In contrast to previous results found by Serrano et al. (2007) in fish, where values for δ13C and δ15N were lower
254
and higher respectively for the farmed group, in our study there is no trend in these values apart from δ 13C for
255
farmed shrimp being out of the range -20 to -14. A potential explanation for these differences between the three
256
farms should be a different diet used to feed the shrimps.
257
The 3D scatterplot of As, Pb and Cd data is represented in Figure 1c. From this plot and the bi-plots in
258
Figure S1, it can be seen that representing Cd vs. As is also helpful for identifying farmed shrimps, since the
259
levels for these trace elements are very low for the group of aquaculture samples. However, no groups could be
260
clearly established in these plots at the origin or species level (Fig. S1 in the supplementary material).
261
When isotopic and/or elemental composition profiles of different origins are clearly different, grouping
262
them into different classes can be achieved with simple methods such as these bi-plots, but in other cases, such
263
as identifying the origin and species levels, the differences are so small that it is necessary to combine a larger
264
number of variables, and, therefore, to apply multivariate statistical techniques. Cluster analysis, PCA, k-means
265
hierarchical clustering and discriminant analysis were applied independently to the different datasets to evaluate
266
whether the methods can correctly classify the samples according to production method (wild or farmed),
267
geographical origin and biological species.
268
9
269
3.1. Wild vs farmed
270
271
Three datasets consisting of more than two variables (As + Cd + Pb analysis, As + Cd + Pb + P + S
272
analysis, or a combination of all five elements with SIR data) were subjected to PCA. When taking and
273
representing three principal components, wild and farmed individuals could be differentiated using the three
274
datasets, although the differentiation is not clear when plotting only the first two principal components using the
275
As + Cd + Pb or the As + Cd + Pb + P + S datasets (Figs. 2a and 2b). But the best differentiation of wild and
276
farmed individuals using PCA was obtained when analysing the dataset with all seven variables coming from
277
both the SIR and the element analyses. In this case, the farmed samples were completely separated from the
278
wild samples in the 2D PCA plots, which is even more evident in the 3D PCA plots (Fig. 2c). An outlier was
279
identified in the datasets, consisting of one of the F. merguiensis samples. The outlier is caused by very high
280
levels of Cd and As for this specimen, while the other variables are at the same level as in the other F.
281
merguiensis samples. Since Cd and As concentrations are considerably higher in the digestive tract
282
(hepatopancreas) than in muscle (Neff, 2002), a potential explanation for these high levels could be a possible
283
contamination of the white muscle used for the analysis with some hepatopancreas residue. The outlier was also
284
evident in the raw data scatterplots (Fig. 1c and Fig. S1 in the supplementary material). If this outlier is
285
eliminated from the PCA analysis, the grouping of samples into wild and farmed individuals becomes clearer
286
(Fig. 2d).
287
The dendrograms generated from cluster analysis of the datasets for C and N isotope ratios; As + Cd +
288
Pb + P + S, and all seven variables showed a certain trend of grouping the two sets of ‘farmed’ shrimps together,
289
although there was not a unique ‘farmed’ cluster completely separated from the ‘wild’ samples (Fig. S2a in the
290
supplementary material). Eliminating the outlier sample identified by scatterplots and PCA didn't yield an
291
improvement in the grouping (results not shown). When an additional fourteen samples were added to the δ13C
292
+ δ15N dataset, no significant improvement in the wild vs. farmed separation was obtained (Fig. S2a). The best
293
grouping of the ‘farmed’ individuals was achieved when analysing only the trace elements dataset (As + Cd +
294
Pb), although not all ‘wild’ samples were grouped together (Fig. S2a in the supplementary material).
295
However, the best results overall were obtained using k-means clustering and, above all, DA (Table 2).
296
Regarding the technique (SIR or multi-element analysis) and the dataset used (δ13C + δ15N for SIR analysis, As
297
+ Cd + Pb + P + S; As + Cd + Pb; or P + S for element analysis, or both together), the highest classification
298
performance was achieved with SIR analysis of δ13C + δ15N (100% for overall performance and 100% for cross-
10
299
validation) and also with the combination of all seven parameters (100% overall performance and 96.9% for
300
cross-validation), both using DA. When using only element analysis, the combination of the five elements (As +
301
Cd + Pb + P + S) performed better than the analysis of only P + S (80.6%) and only As + Cd + Pb (74.2%).
302
When k-means clustering was applied, SIR analysis performed better than SIR and multi-element analysis
303
together (90.3% and 80.6%, respectively). From these results, it may seem that by using SIR analysis of δ13C +
304
δ15N alone, a correct classification of individuals into wild and farmed groups can be achieved, and, therefore,
305
there is no need for multi-element analysis. In fact, some authors have confirmed the influence of diet in the
306
values of δ13C and δ15N in fish muscle (Moreno-Rojas et al., 2008), and SIR analysis has succeeded in
307
differentiating different wild and farmed fish species (Busetto et al., 2008; Dempson et al., 2004; Gordon Bell et
308
al., 2007; Moreno-Rojas et al., 2007; Serrano et al., 2007). Nevertheless, when we increased the sample size
309
with a further fourteen individuals in which only δ13C and δ15N were measured, overall classification decreases
310
to 77.8%, indicating that the combination of SIR with multi-element analysis performs better. This change in the
311
classification rate when increasing the sample size, due to the fact that only four of the added wild samples were
312
correctly classified, highlights the importance of sampling and points out that results from SIR analysis when
313
using a limited number of samples, as many studies do, have to be taken with care. Moreover, when the outlier
314
identified by bi-plots and PCA was eliminated from the statistical analysis of multi-element datasets, overall
315
classification performance was not affected, which indicates that element analysis is able to overcome possible
316
outlier individuals.
317
318
3.2. Origin
319
320
For origin assessment, a total of nine different groups were considered, namely FAO area 71,
321
Argentina, North Atlantic, Mozambique, Nigeria, Senegal, and farms A, B and C.
322
Plotting the content of pairs of elements has been described as a good technique for discriminating between
323
different geographical origins in some food commodities (Geana, Iordache, Ionete, Marinescu, Ranca & Culea,
324
2013). Nevertheless, in the current study, the differences in the elements analysed were small, so no
325
differentiation of the geographical origins could be done plotting only two variables. Multivariate statistical
326
techniques were then applied to a larger number of variables. As raw data scatterplots, PCA analysis barely
327
succeeded in differentiating among these origins within the ‘wild’ group, as shown in the PCA 2D and 3D
328
graphics (Fig. S3 in the supplementary material). Only those samples originating in ‘Senegal’ seemed to be far
11
329
enough from and not mixed with the other ‘wild’ samples when the As + Cd + Pb + P +S dataset was used for
330
the PCA (Fig S3a in the supplementary material). On the other hand, farmed samples from different farms could
331
be differentiated using the complete dataset (Fig. S3c in the supplementary material). Interestingly, PCA,
332
together with DA, is the chemometric tool most frequently used in combination with SIR or multi-element
333
analysis for the discrimination of foods with geographical indications (Gonzalvez et al., 2009).
334
The dendrograms generated from the cluster analysis (Fig. S2b in the supplementary material) showed
335
as some actual groups of samples formed separated clusters. A clear example was the case from the SIR datasets
336
for Argentina samples (Fig. S2b). Cluster analysis also managed to differentiate between the different
337
aquaculture farms, even when samples were from the same species (L. vannamei). Another interesting result
338
was that ‘Senegal’ samples and ‘Mozambique’ samples grouped in different clusters, although mixed with some
339
‘Nigeria’ and ‘Area 71’ individuals. We can conclude that cluster analysis can help, but is not the best approach
340
for assessing origin.
341
The most interesting results were again obtained with DA (Table 3). Stable isotope ratio analysis was
342
able to correctly classify above 70% of the samples, and showed to be independent of sample size (increased
343
from 71.0 to 75.6% when adding new 14 samples), although success rate for element analysis was higher
344
(93.5%). However, the combination of both techniques achieved a 100% correct classification of the samples
345
into eight geographical regions (80% for cross-validation). Although SIR analysis is generally employed more
346
often than multi-element analysis, the capabilities of combining both strategies in order to differentiate
347
geographical origins have been highlighted previously (Gonzalvez et al., 2009). When using only elemental
348
analysis, P and S analysis yielded a much higher overall performance than As + Cd + Pb (80.6% to 51.6%,
349
respectively). The inclusion/exclusion of the outlier from the elemental analysis datasets did not affect the
350
classification figures, indicating the ability of the analysis to overcome possible outliers. It is important to note
351
that all individuals from the three different aquaculture farms were correctly classified by DA independently of
352
the technique or dataset used. The good performance of DA for origin assessment agreed with the figures in
353
previous studies in which meats from different origins were differentiated using this multivariate statistical tool
354
(Piasentier, Valusso, Camin & Versini, 2003; Renou, Bielicki, Deponge, Gachon, Micol & Ritz, 2004).
355
356
3.3. Species authentication
357
12
358
Although SIR analysis in muscle has been shown to be useful for differentiating some fish species by
359
just plotting δ13C vs. δ15N (Monteiro Oliveira et al., 2011), data obtained in the present study revealed that the
360
differences between shrimp species are too small to differentiate them using these bi-plots (Fig. S1 in the
361
supplementary material). Therefore, the datasets obtained were subjected to different multivariate analyses.
362
Principal component analysis didn't provide adequate results for differentiating among the different
363
species within the ‘wild’ group (Fig. S4 in the supplementary material), although there is a certain trend of
364
differentiation from the other samples for the P. muelleri, P. monodon and P. notialis individuals when
365
analysing the five variables from the element analysis (Fig. S4a in the supplementary material). The six L.
366
vannamei samples were well differentiated in the dataset combining the SIR and the element analysis (Fig S4c),
367
but this might be because they all came from aquaculture origin, rather than due to differences at the species
368
level.
369
Cluster analysis was also performed. Although most of the samples belonging to each species are close
370
in the dendrograms generated, most of the clusters are formed by a mix of several species (Fig. S2c in the
371
supplementary material). The two species that formed separate clusters in most of the analyses were P. muelleri
372
and L. vannamei. However, results obtained for species differentiation using cluster analysis of SIR and element
373
data were not as good as those previously reported for DNA (Pascoal, Barros-Velázquez, Cepeda, Gallardo &
374
Calo-Mata, 2008b) and proteomics data (Ortea, Cañas, Calo-Mata, Barros-Velázquez & Gallardo, 2009), where
375
phylogenies could be established.
376
Regarding the use of DA at the species level (Table 4), SIR analysis was able to classify correctly only
377
58.1% of the samples, although rising to 71.1% when the new 14 samples dataset was added. The use of only
378
element analysis increased the global hit to 74.2% when using the five measured elements. Results were poor
379
when limiting the analysis to only P + S or A + Cd + Pb subsets (61.3% and 45.2%, respectively). But the best
380
classification figures were obtained with the whole dataset. Significantly higher classification accuracies
381
(overall 93.5%) were achieved using the combination of SIR analysis and multi-element analysis, and were not
382
influenced by the inclusion or exclusion of the outlier observation (80% in the cross-validation).
383
It has to be noted that, regarding the geographic areas where the different species groups were
384
collected, three of the species, namely F. notialis, P. monodon and L. vannamei, were collected from different
385
areas (F. notialis from two different origins, Senegal and Nigeria; P. monodon from two different origins,
386
Mozambique and farm b; and L. vannamei from farms a and c). For the other species, F. indicus, F.
387
merguiensis, P. muelleri and P. borealis, all specimens in each group were caught in the same general area
13
388
(Mozambique, FAO 71, Argentina and North Atlantic, respectively), although the six F. indicus the nine F.
389
merguiensis and the six P. muelleri specimens were caught in batches of three specimens. Since each batch was
390
caught by fishing vessels at a different time, it can be supposed that they came from not exactly the same
391
geographic place, but different spots within the general fishing area for that species. However, the exact location
392
where each batch was captured was not recorded. Therefore, a certain percentage of the success rate in the
393
species classification could be caused by an artifact of the samples from this four species being collected from
394
the same geographic region, and further studies should be done including several geographic origins for all of
395
the species studied, in order to validate the results related to species classification.
396
397
4. Conclusion
398
399
This study demonstrated the capability of SIR and multi-element analysis with multivariate statistics
400
data handling to classify shrimp samples into different groups according to the method of production (wild or
401
farmed), the geographical origin and the biological species. Although multi-element analysis is more generally
402
used than SIR analysis, we found that the combination of both techniques can enhance the prediction
403
capabilities of chemometric-data analysis in order to classify shrimp samples into wild/farmed, different
404
geographical origins or even biological species. If, for any reason, only one technique can be applied, SIR was
405
able to provide the correct differentiation between wild and farmed, although multi-element analysis worked
406
better for origin and species assessment according to classification rates reported.
407
Regarding the data treatments applied for grouping, DA showed up as the most accurate tool for
408
obtaining the correct classification. Although bi-plots, PCA and cluster analysis can help in differentiating wild
409
and farmed individuals where isotopic and/or elemental composition profiles are usually clearly different, they
410
weren’t appropriate for differentiation at the origin and species levels, where differences are smaller. The
411
discriminant analysis classification results based on the composition of a limited number of elements (As, Cd,
412
Pb, P and S) were generally better than those coming from SIR analysis (δ13C and δ15N), but the use of a
413
combination of all seven parameters from both techniques is recommended, since it yielded a 100% correct
414
prediction for the origin and the wild/farmed groupings, and the highest score for species assessment, without
415
being affected by the presence of possible outliers. Moreover, success rates for wild/farmed classification
416
obtained using only SIR decreased when adding a new set of 14 samples (although for origin classification
417
increased slightly), what points out that reliability of this method should be taken with care when a limited
14
418
number of samples are analysed, even when cross-validation tells us that the results are acceptable for the whole
419
population and not only for the considered sample.
420
To the best of our knowledge, this study represents the first time that SIR and multi-element analyses
421
have been applied at the same time to assess origin, species and production method in shellfish, and we can
422
anticipate that these methodologies will play a role in the evaluation of seafood product authenticity in the
423
future, helping to prevent fraudulent practices in the related industries. In spite of the promising results obtained,
424
further research with a larger number of samples in each group and including several geographic origins for all
425
the species studied, is needed to confirm the validity of these approaches.
426
427
428
Acknowledgments
429
430
We thank Eva María Rodríguez and Lorena Barros for their excellent technical assistance. We also
431
acknowledge members from CETMAR for their helpful collaboration in the collection of specimens for this
432
study.
433
434
435
References
436
437
Anderson, K. A., Hobbie, K. A., & Smith, B. W. (2010). Chemical profiling with modeling differentiates wild
438
and farm-raised salmon. Journal of Agricultural and Food Chemistry, 58, 11768-11774.
439
440
Buchhholz, U., Bernard, H., Werber, D., Böhmer, M. M., Remschmidt, C., Wilking, H., et al. (2011). German
441
outbreak of Escherichia coli O104:H4 associated with sprouts. The New England Journal of Medicine, 365,
442
1763-1770.
443
444
Busetto, M. L., Moretti, V. M., Moreno-Rojas, J. M., Caprino, F., Giani, F., Malandra, R. Bellagamba, F, &
445
Guillou, C. (2008). Authentication of farmed and wild turbot (Psetta maxima) by fatty acid and isotopic analyses
446
combined with chemometrics. Journal of Agricultural and Food Chemistry, 56, 2742-2750.
447
15
448
Council Regulation (EC) No 104/2000 of 17 December (1999) on the common organization of the markets in
449
fishery and aquaculture products. Official Journal of the European Communities, L 17, 22–52.
450
451
Dempson, J.B., & Power, M. (2004). Use of stable isotopes to distinguish farmed from wild Atlantic salmon,
452
Salmo salar. Ecology of Freshwater Fish, 13, 176-184.
453
454
Drivelos, S. A., & Georgiou, C. A. (2012). Multi-element and multi-isotope-ratio analysis to determine the
455
geographical origin of foods in the European Union. Trends in Analytical Chemistry, 40, 38-51.
456
457
FAO Fisheries and Aquaculture Department, Food and Agriculture Organization of the United Nations. (2012).
458
The State of World Fisheries and Aquaculture 2012. FAO Communication Division, Italy: Rome.
459
460
Folch, J., Lees, M., & Stanley, G. H. S. (1957). A simple method for the isolation and purification of total lipids
461
from animal tissues. Journal of Biological Chemistry, 226, 497−509.
462
463
Geana, I., Iordache, A., Ionete, R., Marinescu, A., Ranca, A., & Culea, M. (2013). Geographical origin
464
identification of Romanian wines by ICP-MS elemental analysis. Food Chemistry, 138, 1125-1134.
465
466
Gonzálvez, A., Armenta, S., & de la Guardia, M. (2009). Trace-element composition and stable-isotope ratio for
467
discrimination of foods with Protected Designation of Origin. Trends in Analytical Chemistry, 28, 1295-1311.
468
469
Gordon Bell, J., Preston, T., James Henderson, R., Strachan, F., Bron, J. E., Cooper, K., & Morrison, D. J.
470
(2007). Discrimination of wild and cultured European sea bass (Dicentrarchus labrax) using chemical and
471
isotopic analyses. Journal of Agricultural and Food Chemistry, 55, 5934-5941.
472
473
Kelly, S., Heaton, K., & Hoogewerff, J. (2005). Tracing the geographical origin of food: The application of
474
multi-element and multi-isotope analysis. Trends in Food Science & Technology, 16, 555-567.
475
476
Kim, K.W., & Thornton, Y. (1993). Influence of Ordovician uraniferous black shales on the trace element
477
composition of soils and food crops, Korea. Applied Geochemistry Supplement, 2, 249-255.
16
478
479
Larson, C. (2013). Tense vigil in China as nasty flu virus stirs back to life. Science, 342, 1031.
480
481
Liu, X., Xue,C., Wang, Y., Li, Z., Xue, Y., & Xu, J. (2012). The classification of sea cucumber (Apostichopus
482
japonicus) according to region of origin using multi-element analysis and pattern recognition techniques. Food
483
Control, 23, 522-527.
484
485
Monteiro Oliveira, E. J. V., Sant'Ana, L. S., Ducatti, C., Denadai, J. C., & de Souza Kruliski, C. R. (2011). The
486
use of stable isotopes for authentication of gadoid fish species. European Food Research and Technology, 232,
487
97-101.
488
489
Moreno-Rojas, J.M., Serra, F., Giani, I., Moretti, V.M., Reniero, F., & Guillou, C. (2007). The use of stable
490
isotope ratio analyses to discriminate wild and farmed gilthead sea bream (Sparus aurata). Rapid
491
Communications in Mass Spectrometry, 21, 207-2011.
492
493
Moreno-Rojas, J. M., Tulli, F., Messina, M., Tibaldi, E., & Guillou, C. (2008). Stable isotope ratio analysis as a
494
tool to discriminate between rainbow trout (O. mykiss) fed diets based on plant or fish-meal proteins. Rapid
495
Communications in Mass Spectrometry, 22, 3706-3710.
496
497
Moretti, V.M., Turchini, G.M., Bellagamba, F., & Caprino, F. (2003). Traceability issues in fishery and
498
aquaculture products. Veterinary Research Communications, 27, 497-505.
499
500
Neff, J. M. (2002). Bioaccumulation in Marine Organisms: Effect of Contaminants from Oil Well Produced
501
Water. Elsevier Ltd, The Netherlands: Amsterdam.
502
503
Ortea, I., Cañas, B., Calo-Mata, P., Barros-Velázquez, J., & Gallardo, J. M. (2009) Arginine kinase peptide mass
504
fingerprinting as proteomic approach for species identification and taxonomic analysis of commercially relevant
505
shrimp species. Journal of Agricultural and Food Chemistry, 57, 5665-5672.
506
17
507
Ortea, I., Pascoal, A., Cañas, B., Gallardo, J. M., Barros-Velázquez, J., & Calo-Mata, P. (2012). Food
508
authentication of commercially relevant shrimp species: From classical methods to Foodomics. Electrophoresis,
509
33, 2201-2211.
510
511
Oulhote, Y., Le Bot, B., Deguen, S., & Glorennec, P. (2011). Using and interpreting isotope data for source
512
identification. Trends in analytical Chemistry, 30, 302-312.
513
514
Pascoal, A., Barros-Velázquez, J., Cepeda, A., Gallardo, J. M., & Calo-Mata, P. (2008a). Survey of the
515
authenticity of prawn and shrimp species in commercial food products by PCR-RFLP analysis of a 16S
516
rRNA/tRNAVal mitochondrial region. Food Chemistry, 109, 638–646.
517
518
Pascoal, A., Barros-Velázquez, J., Cepeda, A., Gallardo, J. M., & Calo-Mata, P. (2008b). A polymerase chain
519
reaction-restriction fragment length polymorphism method based on the analysis of a 16S rRNA/tRNA(Val)
520
mitochondrial region for species identification of commercial penaeid shrimps (Crustacea: Decapoda:
521
Penaeoidea) of food interest. Electrophoresis, 29, 499-509.
522
523
Piasentier, E., Valusso, R., Camin, F., & Versini, G. (2003). Stable isotope ratio analysis for authentication of
524
lamb meat. Meat Science, 64, 239-247.
525
526
Quinn, B. (2013). Horse DNA found in beefburgers from four major supermarkets, The Guardian 16 January
527
2013. Available at http://www.theguardian.com/world/2013/jan/15/horse-dna-found-supermarket-beefburgers,
528
accessed date April 2014.
529
530
Rasmussen, R. S., & Morrisey, M. T. (2008). DNA-based methods for the identification of commercial fish and
531
seafood species. Comprehensive Reviews in Food Science and Food Safety, 7, 280-295.
532
533
Regulation (EC) No 178/2002 of the European Parliament and of the Council of 28 January (2002) laying down
534
the general principles and requirements of food law, establishing the European Food Safety Authority and
535
laying down procedures in matters of food safety. Official Journal of the European Communities, L31, 1-24.
536
18
537
Renou, J-P., Bielicki, G., Deponge, C., Gachon, P., Micol, D., & Ritz, P. (2004). Characterization of animal
538
products according to geographic origin and feeding diet using nuclear magnetic resonance and isotope ratio
539
mass spectrometry. Part II: Beef meat. Food Chemistry, 86, 251-256.
540
541
Serrano, R., Blanes, M. A., & Orero, L. (2007). Stable isotope determination in wild and farmed gilthead sea
542
bream (Sparus aurata) tissues from the western Mediterranean. Chemosphere, 69, 1075-1080.
543
544
Smith, R. G. & Watts, C. A. (2009). Determination of the country of origin of farm-raised shrimp (family
545
penaeide) using trace metal profiling and multivariate statistics. Journal of Agricultural and Food Chemistry,
546
57, 8244–8249.
547
548
Spink, J., & Moyer, D.C. (2011). Defining the public health threat of food fraud. Journal of Food Science, 76,
549
R157-R163.
550
551
U.S. Food and Drug Administration. (2013). US Federal Food, Drug, and Cosmetic Act. Sec. 403. Misbranded
552
food. Available at http://www.fda.gov/RegulatoryInformation/Legislation/FederalFoodDrugand
553
CosmeticActFDCAct/FDCActChapterIVFood/default.htm, accessed date Sep 2013.
554
555
Vinci, G., Preti, R., Tieri, A., & Vieri, S. (2013). Authenticity and quality of animal origin food investigated by
556
stable-isotope ratio analysis. Journal of the Science of Food and Agriculture, 93, 439-448.
557
558
Will, R. G., Ironside, J. W., Zeidler, M., Cousens, S. N., Estibeiro, K., Alperovitch, A., et al. (1996). A new
559
variant of Creutzfeldt-Jakob disease in the UK. The Lancet, 347, 921-925.
560
561
562
563
564
565
566
19
567
Figure captions
568
569
Fig. 1. Scatterplots of (a) C and N isotope ratio analysis for production method, (b) origin assessment and (c)
570
As, Pb and Cd element analysis for production method assessment
571
572
Fig. 2. PCA plots for wild/farmed differentiation using either the first two or the first three principal components
573
for (a) the As + Cd + Pb dataset, (b) the As + Cd + Pb + P + S dataset, (c) the complete dataset, and (d) the
574
complete dataset eliminating the outlier from the analysis
575
576
577
Electronic supplementary material
578
579
Fig. S1. Scatterplots of (a) δ13C vs. δ15N; (b) As vs. Cd; (c) As vs. Pb; (d) Pb vs. Cd; (e) As vs. Pb vs. Cd ; and
580
(f) P vs. S data obtained from SIR and multielement analysis in shrimp samples according to production
581
method, origin and biological species
582
583
Fig. S2. Dendrograms generated from the cluster analysis of the different datasets according to (a) production
584
method; (b) origin; and (c) biological species
585
586
Fig. S3. PCA plots for geographical origin analysis representing the two and three principal components for (a)
587
the As+Cd+Pb+P+S dataset; (b) the As+Cd+Pb dataset; and (c) the complete dataset (As + Cd + Pb + P + S +
588
δ13C + δ15N). 31 samples were included in all the three datasets
589
590
Fig. S4. PCA plots for biological species representing the two and three principal components for (a) the As +
591
Cd + Pb + P + S dataset; (b) the As + Cd + Pb dataset; and (c) the complete dataset (As + Cd + Pb + P + S +
592
δ13C + δ15N). 31 samples were included in all the three datasets
593
594
Table S1. Species, geographic origin, batch and production method of every specimen analysed
595
596
20
Table 1. Samples and experimental results of SIR analysis of C and N, and elements measured, according to production method, geographical origin and species.
SIR analysis
δ13C (‰)
mean
Production method
farmed (n = 9)
Wild (n = 36)
Geographical origin
SD
Min
Element analysis
δ15N (‰)
Max
mean SD
Min
As (mg(kg)
Max
-14.88
-16.74
4.80 -21.35 -10.38
0.89 -19.50 -15.22
8.96 3.50 4.49
11.09 2.79 7.95
13.55
17.36
FAO 71 (n = 9)
Argentina (n = 6)
North Atlantic (n = 3)
Farm A (n = 3)
Farm B (n = 3)
Farm C (n = 3)
Mozambique (n = 9)
Nigeria (n = 3)
Senegal (n = 6)
-16.26
-16.12
-18.07
-12.40
-21.21
-11.04
-16.77
-17.87
-16.80
0.54
0.12
0.16
0.26
0.22
0.71
0.37
1.65
1.01
-16.93
-16.34
-18.25
-12.65
-21.35
-11.79
-17.23
-19.50
-17.97
-15.46
-15.99
-17.93
-12.13
-20.95
-10.38
-16.19
-16.20
-15.22
9.71
16.86
10.78
12.92
9.08
4.88
10.55
10.34
8.72
1.07
0.51
0.44
0.56
0.06
0.35
0.53
0.62
0.83
8.56
15.92
10.49
12.51
9.02
4.49
9.52
9.63
7.95
11.66
17.36
11.28
13.55
9.15
5.15
11.46
10.80
10.31
Species
F. indicus (n = 6)
F. merguiensis (n = 9)
F. notialis (n = 9)
L. vannamei (n = 6)
P. borealis (n = 3)
P. monodon (n = 6)
P. muelleri (n = 6)
-16.87
-16.26
-17.15
-11.72
-18.07
-18.90
-16.12
0.37
0.54
1.27
0.88
0.16
2.54
0.12
-17.23
-16.93
-19.50
-12.65
-18.25
-21.35
-16.34
-16.19
-15.46
-15.22
-10.38
-17.93
-16.34
-15.99
10.70
9.71
9.26
8.90
10.78
9.66
16.86
0.45
1.07
1.09
4.42
0.44
0.76
0.51
10.09
8.56
7.95
4.49
10.49
9.02
15.92
11.46
11.66
10.80
13.55
11.28
10.62
17.36
mean SD
Production method
Farmed (n = 6)
Wild (n = 25)
Geographical origin
FAO 71 (n = 6)
Argentina (n = 3)
North Atlantic(n = 1)
Farm A (n = 3)
Farm C (n = 3)
Mozambique (n = 6)
Nigeria (n = 3)
Senegal (n = 6)
Species
F. indicus (n = 3)
F. merguiensis (n = 6)
F. notialis (n = 9)
L. vannamei (n = 6)
P. borealis (n = 1)
P. monodon (n = 3)
P. muelleri (n = 3)
Min
Cd (mg/kg)
Max
mean SD
Min Max
Pb (mg/kg)
mean SD Min Max
mean
SD
P (mg/kg)
Min
Max
mean
SD
S (mg/kg)
Min
Max
4.41 0.87 2.89 5.36
44.75 40.88 9.19 182.26
0.01
0.09
0.01 0.00 0.02
0.11 0.02 0.59
0.05
0.04
0.01 0.04 0.07
0.01 0.03 0.08
9903
11734
2776
1412
6981
8973
12967
13958
8406
11075
1560
1820
6838
8456
10491
15978
77.64
13.70
17.42
4.39
4.44
57.14
46.83
18.51
58.84
4.61
na
0.27
1.35
35.45
24.67
3.20
30.83
9.19
17.42
4.17
2.89
20.44
18.35
14.67
182.26
18.41
17.42
4.69
5.36
100.12
61.76
23.82
0.19
0.11
0.05
0.02
0.01
0.06
0.04
0.06
0.20
0.03
na
0.00
0.00
0.03
0.03
0.03
0.05
0.07
0.05
0.02
0.00
0.02
0.02
0.02
0.59
0.12
0.05
0.02
0.01
0.11
0.07
0.10
0.04
0.04
0.05
0.04
0.06
0.04
0.04
0.04
0.01
0.01
na
0.00
0.00
0.02
0.01
0.01
0.03
0.04
0.05
0.04
0.06
0.03
0.03
0.03
0.06
0.05
0.05
0.04
0.07
0.08
0.04
0.05
12490
12664
9390
7401
12404
12897
11600
9807
264
307
na
383
584
759
529
612
11984
12431
9390
6981
11802
12055
10994
8973
12709
13012
9390
7730
12967
13958
11969
10380
11778
13383
8456
7054
9757
9392
10973
11389
2302
1038
na
271
730
757
764
822
9732
12423
8456
6838
9032
8463
10199
10405
15978
14486
8456
7358
10491
10319
11726
12494
25.42
77.64
27.95
4.41
17.42
88.86
13.70
4.56
58.84
18.95
0.87
na
10.09
4.61
20.44
30.83
14.67
2.89
17.42
80.63
9.19
29.37
182.26
61.76
5.36
17.42
100.12
18.41
0.05
0.19
0.06
0.01
0.05
0.07
0.11
0.02
0.20
0.03
0.01
n
0.04
0.03
0.03
0.05
0.02
0.00
0.05
0.02
0.07
0.06
0.59
0.10
0.02
0.05
0.11
0.12
0.04
0.04
0.04
0.05
0.05
0.05
0.04
0.01
0.01
0.01
0.01
na
0.03
0.01
0.03
0.03
0.03
0.04
0.05
0.03
0.04
0.04
0.06
0.05
0.07
0.05
0.08
0.05
12889
12490
10404
9903
9390
12904
12664
757
264
1053
2776
na
931
307
12055
11984
8973
6981
9390
12194
12431
13532
12709
11968
12967
9390
13958
13012
10026
10404
11250
8406
8456
8759
13383
346
1053
782
1560
na
333
1038
9644
8973
10199
6838
8456
8463
12423
10319
11968
12494
10491
8456
9119
14486
21
Discriminant analysis
k-means classification
Table 2. Performance of the k-means and discriminant analysis classification methods for authenticating
method of production (wild vs. farmed) depending on the dataset used.
SIRa (N=31)
ICP-MSb
ICP-MSb (excluding outlier)
ICP-MS (As, Cd, Pb)
ICP-MS (P+S)
SIRa + ICP-MSb
wild
25/25
22/25
21/24
8/25
22/25
22/25
Correctly classified/total
farmed
3/6
3/6
3/6
6/6
3/6
3/6
Total
28/31
25/31
24/30
14/31
25/31
25/31
SIRa + ICP-MSb (excluding outlier)
21/24
3/6
24/30
80%
SIRa (N=45)
30/36
3/9
33/45
73.3%
SIRa (N=31)
25/25
6/6
31/31
100% (100%)
ICP-MSb
ICP-MSb (excluding outlier)
ICP-MS (As, Cd, Pb)
ICP-MS (P+S)
SIRa + ICP-MSb
SIRa + ICP-MSb (excluding outlier)
22/25
6/6
28/31
90.3% (87.1%)
22/24
19/25
21/25
25/25
24/24
6/6
4/6
4/6
6/6
6/6
28/30
23/31
25/31
31/31
30/30
93.3% (90%)
74.2% (74.2%)
80.6% (74.2%)
100% (96.9%)
100% (96.8%)
SIRa (N=45)
29/36
6/9
35/45
77.8% (75.6%)
Overall (cross validation)
90.3%
80.6%
80%
45.2%
80.6%
80.6%
SIR (stable-isotope ratio) analysis: δ13C and δ15N.
b
ICP-MS multielement analysis: As, Cd, Pb, P and S.
a
22
Table 3. Results from the classification matrices for origin prediction applying discriminant analysis to the different datasets.
Correctly predicted/total
SIRa (N=31)
ICP-MSb
FAO
area
71
1/6
5/6
Argentina
North
Atlantic
Farm A
Farm B
Farm C
Mozambique
Nigeria
Senegal
Total
Overall (crossvalidation)
3/3
3/3
1/1
1/1
3/3
3/3
-
3/3
3/3
5/6
5/6
2/3
3/3
4/6
6/6
22/31
29/31
71.0% (63.3%c)
93.5% (70%c)
ICP-MSb (excluding outlier)
4/5
3/3
1/1
3/3
-
3/3
5/6
3/3
6/6
28/30
93.3% (79.3%c)
ICP-MS (As, Cd, Pb)
ICP-MS (P+S)
SIRa + ICP-MSb
SIRa + ICP-MSb (excluding outlier)
SIRa (N=45)
1/6
3/6
6/6
5/5
4/9
2/3
3/3
3/3
3/3
6/6
1/1
1/1
1/1
1/1
2/3
3/3
3/3
3/3
3/3
3/3
3/3
3/3
2/3
3/3
3/3
3/3
1/6
4/6
6/6
6/6
8/9
2/3
3/3
3/3
3/3
1/3
3/6
6/6
6/6
6/6
4/6
16/31
25/31
31/31
30/30
29/45
51.6% (33.3%c)
80.6% (63.3%c)
100% (80%c)
100% (82.8%c)
75.6% (66.7%)
SIR (stable-isotope ratio) analysis: δ13C and δ15N.
ICP-MS multielement analysis: As, Cd, Pb, P and S.
c
cross validation data excluding the only P. borealis sample
a
b
23
Table 4. Results from the classification matrices for species prediction applying discriminant analysis on the different datasets.
Correctly predicted/total
L. vannamei
P. borealis
6/6
1/1
F. indicus
1/3
F. merguiensis
3/6
F. notialis
3/9
ICP-MSb
3/3
3/6
7/9
3/6
b
3/3
2/3
3/3
3/3
3/3
6/6
3/5
1/6
3/6
6/6
5/5
5/9
7/9
3/9
7/9
7/9
7/9
3/9
6/6
1/6
0/6
6/6
6/6
6/6
SIRa (N=31)
ICP-MS (excluding outlier)
ICP-MS (As, Cd, Pb)
ICP-MS (P+S)
SIRa + ICP-MSb
SIRa + ICP-MSb (excluding outlier)
SIRa (N=45)
Overall (cross-validation)
P. monodon
1/3
P. muelleri
3/3
Total
18/31
1/1
3/3
3/3
23/31
74.2% (66.7%c)
1/1
1/1
1/1
1/1
1/1
3/3
3/3
3/3
3/3
3/3
3/3
3/6
3/3
3/3
2/3
3/3
3/3
6/6
26/30
14/31
19/31
29/31
28/30
32/45
86.7% (72.4%c)
45.2% (33.3%c)
61.3% (66.7%c)
93.5% (80%c)
93.3% (79.3%c)
71.1% (66.7%)
58.1% (60%c)
SIR (stable-isotope ratio) analysis: δ13C and δ15N.
ICP-MS multielement analysis: As, Cd, Pb, P and S.
c
cross validation data excluding the only P. borealis sample
a
b
24
Fig. 1
25
Fig. 2
26