Food Chemistry 136 (2013) 1557–1561

Contents lists available at SciVerse ScienceDirect

Food Chemistry

journal homepage: www.elsevier .com/locate / foodchem

The application of Near-Infrared Reflectance Spectroscopy (NIRS) to detectmelamine adulteration of soya bean meal

Simon A. Haughey a,⇑, Stewart F. Graham a, Emmanuelle Cancouët b, Christopher T. Elliott a

a Institute of Agri-Food and Land Use, School of Biological Sciences, Queen’s University, David Keir Building, Stranmillis Road, Belfast, Northern Ireland BT9 5AG, United Kingdomb UFR Sciences et Techniques, Faculté des Sciences de Nantes, 2 rue de la Houssinière, BP 32229, 44 322 Nantes Cedex 3, France

a r t i c l e i n f o

Article history:Available online 1 February 2012

Keywords:Near Infrared Reflectance SpectroscopyAdulterationAnimal feedSoyaMelamine

0308-8146/$ - see front matter � 2012 Elsevier Ltd. Adoi:10.1016/j.foodchem.2012.01.068

Abbreviations: NIRS, Near Infrared Reflectance SLinked Immunosorbent Assays; FPIA, Fluorescence PoSurface Plasmon Resonance; GC–MS, Gas-ChromatoHPLC, High Performance Liquid Chromatography; TSpectrometry; PCs, principal components; PCA, principrincipal component regression; PLS, partial least sqvariate; RMSEC, root mean squared error of calibratioerror of prediction.⇑ Corresponding author. Tel.: +44 (0) 2890976525;

E-mail address: s.a.haughey@qub.ac.uk (S.A. Haug

a b s t r a c t

Soya bean products are used widely in the animal feed industry as a protein based feed ingredient andhave been found to be adulterated with melamine. This was highlighted in the Chinese scandal of2008. Dehulled soya (GM and non-GM), soya hulls and toasted soya were contaminated with melamineand spectra were generated using Near Infrared Reflectance Spectroscopy (NIRS). By applying chemomet-rics to the spectral data, excellent calibration models and prediction statistics were obtained. The coeffi-cients of determination (R2) were found to be 0.89–0.99 depending on the mathematical algorithm used,the data pre-processing applied and the sample type used. The corresponding values for the root meansquare error of calibration and prediction were found to be 0.081–0.276% and 0.134–0.368%, respectively,again depending on the chemometric treatment applied to the data and sample type. In addition, adopt-ing a qualitative approach with the spectral data and applying PCA, it was possible to discriminatebetween the four samples types and also, by generation of Cooman’s plots, possible to distinguishbetween adulterated and non-adulterated samples.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction food, animal feed and feed materials, wheat gluten, and other pro-

Melamine, 1,3,5-triazine-2,4,6-triamine, is a trimer of cyana-mide and its chemical structure is shown in Fig. 1. It is easilypolymerised through the reaction with formaldehyde to form aheat-tolerant, fire-resistant resin used in the manufacture of flameretardant materials. Its durability and ease of use means that inthis polymer form it has been used for decades in the manufactur-ing of dishes, plastic resins, flame-retardant fibres, components ofpaper and paperboard, industrial coatings, whiteboards, flooringand kitchen utensils (Andersen et al., 2008; Garber, 2008). How-ever, in 2007–2008, melamine gained substantial notoriety whenit was used to intentionally adulterate milk, food and feed materi-als from China, thus fraudently increasing the apparent proteincontent of the products as standard protein determination assays(e.g. Kjeldahl) cannot differentiate between protein nitrogen andnon-protein nitrogen. Residues of melamine were detected in pet

ll rights reserved.

pectroscopy; ELISA, Enzymelarisation Immunoassay; SPR,graphy Mass-Spectrometry;

OF-MS, Time of Flight Masspal component analysis; PCR,uares; SNV, standard normaln; RMSEP, root mean squared

fax: +44 (0) 2890976513.hey).

tein-based food commodities. As a result pet illness and death wassubsequently widely reported (Andersen et al., 2008). In the 2008,several companies and individuals in China were implicated in ascandal involving milk and infant formula which had been adulter-ated with melamine, leading to kidney stones, renal failure, andurinary tract effects in approximately 300,000 children with sixreported deaths (Brown et al., 2007; Ehling, Tefera, & Ho, 2007;Gossner et al., 2009; Yan, Zhou, Zhu, & Chen, 2009). Due to thisand the subsequent discovery of melamine in soya, soya productsand ammonium bicarbonate, the European Commission bannedthe imports of many foodstuffs from China and set the maximumpermitted concentration for melamine in food at 2.5 mg kg�1

(Commission Decision 2008/757/EC, 2008a; Commission Decision2008/798/EC, 2008b; Commission Decision 2008/921/EC, 2008c).Reviews have been carried on the melamine adulteration scandal,including how it unfolded and the worldwide consequences gener-ated in its aftermath (Sharma & Paradakar, 2010) and its implica-tions for food safety regulations (Pei et al., 2011).

Melamine is not permitted to be directly added to foods orfeeds, nor is it permitted to be used as a fertilizer worldwide. Asa result of the discovery of melamine being used to adulterate foodand feed products, the requirement for rapid detection methods formelamine adulteration in these commodities has become animportant part of feed and food routine monitoring schemes. Sincethe melamine crisis of 2007–2008, there have been a number oftests developed on different platforms to detect melamine at low

Fig. 1. Structure of melamine.

1558 S.A. Haughey et al. / Food Chemistry 136 (2013) 1557–1561

levels. These include Enzyme Linked Immunosorbent Assays (ELISA)for commodities such as milk, milk powders, wheat products andfeeds (Garber & Brewer, 2010; Lei et al., 2010, 2011; Yin et al.,2010); Fluorescence Polarisation Immunoassay (FPIA) for milkand milk powders (Wang et al., 2011); Surface Plasmon Resonance(SPR) biosensor assay for infant formula (Fodey et al., 2011); Gas-Chromatography Mass-Spectrometry (GC–MS) for milk powder(Miao et al., 2009); High Performance Liquid Chromatography(HPLC) for infant formula (Venkatasami & Sowa, 2010); Time ofFlight Mass Spectrometry (TOF-MS) for milk powders (Vaclavik,Rosmus, Popping, & Hajslova, 2010). Many of the traditional andnovel detection methods have been recently reviewed (Lin, 2009).

As an alternative to the aforementioned methods, Near-InfraredReflectance Spectroscopy (NIRS) is a technique that has been ap-plied to a wide range of studies involving food and feed analysisincluding areas such as nutrition, authenticity and traceability offood stuffs in the agricultural sector. NIRS is rapidly becoming animportant tool of analysis due to the speed, reproducibility, non-destructive capabilities and in particular due to the relative easeof implementing this low-cost technology into an industrial set-ting, i.e. as an on-line or in-line application. NIRS applications inthe feed and food sectors include its implementation to investigatethe chemical and physical characteristics of wheat as a means ofpredicting the nutritive value of grain for broiler chickens (Owens,McCann, McCracken, & Park, 2009) and to predict the fatty acidcomposition of beef to include the saturated, branched and mono-unsaturated fatty acids (Sierra et al., 2008) and NIRS is very muchan accepted tool in the animal feed sector. There have been manymethods, including NIRS (Balabin & Smirnov, 2011; Mauer,Chernyshova, Hiatt, Deering, & Davis, 2009), that have been devel-oped for the detection of melamine in milk, milk powders andinfant formula. However, there remains a lack of methods appliedfor the detection of melamine in soya and soya products. Severalinternational analytical laboratories and instrumentation provid-ers offer their in-house services or application notes to detectmelamine in soya and soya products using GC–MS and LC–MS

Fig. 2. NIRS spectra of melamine, s

techniques which are very sensitive yet expensive techniqueswhich tend to only be available in large analytical laboratories.

In the present study, the main objective was to develop NIRScalibration models capable of qualitatively and quantitatively pre-dicting levels of melamine adulteration in soya and soya productsamples destined for animal feed production.

2. Materials and methods

2.1. Sample collection and preparation

Batches of commercial soya bean meal samples (hulls, dehulledand toasted) were obtained from the animal feed producing com-pany, John Thompson and Sons Ltd. (Belfast, Northern Ireland).Samples were ground using an analytical mill (IKA Werke,Germany) and subsequently spiked with a range of concentrations(0%, 0.1%, 0.25%, 0.5%, 0.75%, 1%, 1.25%, 1.5%, 1.75% and 2%) ofmelamine (Sigma–Aldrich, UK) used to produce the quantitativecalibration models.

2.2. Near-Infrared Reflectance Spectroscopy

Near-infrared reflectance spectra were recorded on an Antaris IIFT-NIR (Thermo Fisher Scientific, Dublin, Ireland). The spiked soyasamples (10 g) were poured onto the sample cup spinner on theIntegrating Sphere module of the instrument. All the spectra werecomputed at 8 cm�1 resolution across the spectral range 12,000–3800 cm�1 and ran in triplicate. Instrument control and initialspectral manipulation were preformed with Result Integrationsoftware (Thermo Fisher Scientific, Dublin, Ireland). The spectrawere recorded at ambient temperature and a total of 64 scans wereacquired for each spectrum.

2.3. Chemometric analysis

Multivariate analysis was used for quantitative and qualitativeanalysis of the NIRS spectroscopic data. The analysis was carriedout using the software package TQ Analyst (Thermo FisherScientific, Dublin, Ireland) and Simca P + 12 (Umetrics, Sweden).Principal components analysis (PCA) and Cooman’s plots, whichare well suited for differentiating feed that contain varyingamounts of the same major components, were used as qualitativetests in this investigation. Principal components regression (PCR)and partial least squares (PLS) were used in the current study for

oya and 2% melamine in soya.

-30

-20

-10

0

10

20

30

-80 -60 -40 -20 0 20 40 60 80 100

PC

2

PC1

Fig. 3. PCA scores plot displaying de-hulled soya (non-GM; red dots), de-hulledsoya (GM; black squares), soya hulls (blue diamond) and toasted soya (greentriangle). Principal component 1 explains 84.9% of the variation whilst PC2 explains14.9%. UV scaled raw data using mean values (n = 3). (For interpretation of thereferences to colour in this figure legend, the reader is referred to the web version ofthis paper.)

S.A. Haughey et al. / Food Chemistry 136 (2013) 1557–1561 1559

quantitative model building. For the quantitative tests the condi-tions used to generate each calibration model was based on obtain-ing the lowest standard error of prediction to minimise the risk ofover-fitting when the model accuracy was evaluated. Models weredeveloped using the wavenumber range 9000–4000 cm�1. Thedata pre-treatments examined were none (raw spectral data), 1stand 2nd derivatives calculated using the Savitzky-Golay method.

3. Results and discussion

3.1. Qualitative analysis of data

Fig. 2 outlines the NIRS spectra of 100% soya, 2% melamine insoya and 100% melamine. It can be seen that a distinct melaminepeak is observed at �6800 cm�1 in the 2% melamine in soya spec-trum and as the concentration present decreased, the visibility ofthe peak reduced (data not shown). As a result the spectral datais analysed using chemometrics; the region between 9000 and4000 cm�1 was selected as the area for analysis. Fig. 3 shows thePCA scores plot for non-GM de-hulled soya, GM de-hulled soya,soya hulls and toasted soya feed materials using the raw data gen-

0

20

40

60

80

100

120

140

160

180

0 20 40 60 80 100 120 140 160 180 200 220 240 260

Dis

tan

ce t

o D

e-hu

lled

Soya

(C

ombi

ned)

D-Crit(0.05)

D-Crit(0.05)

a

Distance to Soya Hulls

Fig. 4. (a) Cooman’s prediction plot showing the combined de-hulled series including thaxis (green squares). Blue triangles correspond to the prediction set spiked with various pprediction set spiked with various percentages of melamine used for soya hulls. Analy4000 cm�1, Savitzky-Golay smoothing was applied along with SNV. (b) Cooman’s predhorizontal axis (green square). Blue triangles correspond to the prediction set spiked withthe prediction set spiked with various percentages of melamine used for toasted soya. A4000 cm�1, Savitzky-Golay smoothing was applied along with SNV. (For interpretationversion of this paper.)

erated which has been scaled using the univariate scaling tech-nique. The first two principal components (PCs), out of the threePCs generated in the PCA calibration model, explain the majorityof the variation with PC1 explaining 84.9% and PC2 explaining14.9%. PCA is an unbiased qualitative technique and the scores plotshows the distinctly separate clusters of the four soya bean prod-ucts including differentiating between GM and non-GM de-hulledsoya.

Cooman’s plots can be used to generate qualitative calibrationmodels using data which includes analyte positive samples of vary-ing concentrations. Examples of these plots are shown in Fig. 4aand b. Fig. 4a displays the Cooman’s plots of the combined de-hulled soya samples (GM and non-GM) versus soya hulls andFig. 4b displays the Cooman plots of the soya hulls versus toastedsoya samples. The plots were generated using distance modelling(DModx+, (Simca-P + 12, Umetrics)) to determine if it could beused as a screening method based on the NIRS data. The plots alsoinclude prediction sets of samples, spiked at various percentageswith melamine, applied to the model. The plot was generated fromcalibration models using the 1st order derivative of the spectraldata in the 9000–4000 cm�1 region, with Savitzky-Golay smooth-ing and standard normal variate (SNV) processing which effec-tively removes the multiplicative interferences of scatter andparticle size. These are useful plots as they can uncover moderateoutliers in the data, i.e. samples whose variable signatures are dif-ferent from the majority of recorded observations. For both plots itwas determined that no outliers existed when unadulterated sam-ples (soya hulls and de-hulled soya or toasted soya) were used. Thespiked samples were contaminated at levels of 0.1–2% melamine.The sample with the highest concentration of melamine was foundto lie furthest from its respective axis. Using these plots, adultera-tion of samples with melamine could be measured with NIRS aslow as 0.1% with a 95% confidence limit (indicated by the red lineson the plots). These results demonstrate that Cooman’s plot mod-elling has the potential to be implemented into an industrialsetting as a means of quality control, i.e. samples could be screenedand therefore passed or failed on-line prior to use in animal feedproduction.

3.2. Quantitative calibration modelling and validation

Quantitative calibration models were generated using PLS andPCR mathematical algorithms which convert complex spectral datainto analytical parameters. Table 1 highlights the results for mela-

0

510

1520

2530

3540

4550

5560

0 10 20 30 40 50 60 70 80 90 100110120130140150

Dis

tanc

e to

Soy

a H

ulls

D-Crit(0.05)

D-Crit(0.05)

b

DistancetoToastedSoya

e GM and non-GM samples (black triangles) versus the soya hulls on the horizontalercentages of melamine used for the de-hulled soya. Red squares correspond to thesis was carried out on the 1st derivative of the spectral data between 9000 and

iction plot showing the soya hulls (black triangle) versus the toasted soya on thevarious percentages of melamine used for the soya hulls. Red squares correspond to

nalysis was carried out on the 1st derivative of the spectral data between 9000 andof the references to colour in this figure legend, the reader is referred to the web

Table 1PLS and PCR calibration models from (i) de-hulled soya (non-GM) adulterated with melamine: 792 spectra were used to produce the calibration and 396 spectra were used in thevalidation of the results; (ii) de-hulled soya (GM) adulterated with melamine: 540 spectra were used to produce the calibration and 270 spectra were used in the validation of theresults; (iii) soya hulls adulterated with melamine: 486 spectra were used to produce the calibration and 283 spectra were used in the validation of the results; (iv) toasted soyaadulterated with melamine: 125 spectra were used to produce the calibration and 64 spectra were used in the validation of the results. For 1st order derivative processing of data,Savitzky-Golay smoothing was applied with data points = 7 and polynomial order = 2. For 2nd order derivative processing of data, Savitzky-Golay smoothing was applied withdata points = 9 and polynomial order = 3.

De-hulled soya (non-GM) De-hulled soya (GM) Soya hulls Toasted soya

Rawdata

1stDerivative

2ndDerivative

Rawdata

1stDerivative

2ndDerivative

Rawdata

1stDerivative

2ndDerivative

Rawdata

1stDerivative

2ndDerivative

PLS SNV R2 0.91098 0.95490 0.96320 0.98684 0.98216 0.98211 0.96548 0.94614 0.93909 0.94395 0.99156 0.99594RMSEC (%) 0.258 0.186 0.169 0.102 0.117 0.170 0.163 0.202 0.215 0.206 0.0810 0.0593RMSEP (%) 0.363 0.271 0.288 0.134 0.143 0.151 0.218 0.251 0.290 0.268 0.135 0.158Factors 6 10 9 8 6 4 10 7 4 2 10 7

PCR SNV R2 0.91018 0.89723 0.89857 0.98453 0.98116 0.98010 0.92370 0.94078 0.92183 0.96053 0.96515 0.97492RMSEC (%) 0.259 0.276 0.276 0.109 0.120 0.124 0.239 0.212 0.242 0.174 0.164 0.139RMSEP (%) 0.362 0.368 0.368 0.139 0.143 0.153 0.289 0.261 0.298 0.203 0.203 0.210PCs 10 10 10 10 10 10 10 10 10 10 10 10

Fig. 5. Graphical representation of the quantitative calibration models generated for melamine adulteration of soya products.

1560 S.A. Haughey et al. / Food Chemistry 136 (2013) 1557–1561

mine adulterated samples of de-hulled soya (GM & non-GM), soyahulls and toasted soya. The data generated for each respective cal-ibration models using the mathematical treatments of the NIRSspectral data includes the corresponding root mean squared errorof calibration (RMSEC) and root mean squared error of prediction(RMSEP). These latter values gave an indication of the quality ofthe calibration models generated. The calibration models for eachset of samples were very good with correlation values (R2) between0.89 and 0.99 depending on the mathematical algorithm used, thedata pre-processing applied and the sample type. The correspond-ing values for RMSEC and RMSEP were found to be 0.081–0.276%and 0.134–0.368%, respectively, again depending on the chemo-metric treatment of the data and sample type. Fig. 5 shows thegraphical representation for each of the best calibration modelsgenerated for each soya sample type with actual concentrationon X-axis versus calculated (predicted) concentration on the Y-axis.

The samples used for calibration have been indicated by the circleswhilst those used for validation are indicated by crosses. The vali-dation results for the models showed variation depending on themathematical treatment applied. In general, the data generatedusing the PLS algorithm gave marginally better calibration models,with R2 closer to 1, and validation results with RMSEP, closer to 0,than when applying the PCR algorithm.

4. Conclusion

The calibration models produced in this study, with excellentcoefficients of determination (R2 = 0.89–0.99) and the subsequentvalidation data generated, RMSEC (0.081–0.276%) and RMSEP(0.134–0.368%), indicate that this NIRS approach has the potentialto be used to detect fraud and adulteration of soya based productsused as animal feed ingredients. Routine testing of shipments of

S.A. Haughey et al. / Food Chemistry 136 (2013) 1557–1561 1561

soya based products with NIRS could have economic benefits forthe animal feed sector and reduce the occurrence of crises likethose mentioned in the introduction. Further studies on these cal-ibration models will include the transfer of these lab-based meth-ods into an industrial setting where the incoming batches of feedmaterials could be screened for fraudulent activity leading to saferanimal feed, healthier animals and ultimately safer food.

Acknowledgements

The authors are very grateful for the supply of soya based sam-ple sets used in this study which were kindly donated by JohnThompson and Sons Ltd., Belfast and also to Anne-Julie Noblet forhelp in preparing and running samples on the NIRS instrument.The work was funded through the Fortress Ireland project by InvestNorthern Ireland and by the European Commission 7th FrameworkProject; Quality and Safety of Feeds and Food for Europe, QSAFFE(FP7-KBBE-2010-4, 265702).

References

Andersen, W. C., Turnipseed, S. B., Karbiwnyk, C. M., Clark, S. B., Madson, M. R.,Gieseker, C. M., et al. (2008). Determination and confirmation of melamineresidues in catfish, trout, tilapia, salmon, and shrimp by liquid chromatographywith tandem mass spectrometry. Journal of Agricultural and Food Chemistry, 56,4340–4347.

Balabin, R. M., & Smirnov, S. V. (2011). Melamine detection by mid- and near-infrared (MIR/NIR) spectroscopy: A quick and sensitive method for dairyproducts analysis including liquid milk, infant formula, and milk powder.Talanta, 85, 562–568.

Brown, C. A., Jeong, K. S., Poppenga, R. H., Puschner, B., Miller, D. M., Ellis, A. E., et al.(2007). Outbreaks of renal failure associated with melamine and cyanuric acidin dogs and cats in 2004 and 2007. Journal of Veterinary Diagnostic Investigation,19, 525–531.

Commission Decision 2008/757/EC imposing special conditions governing theimport of products containing milk or milk products originating in or consignedfrom China (2008a). Official Journal of the European Union, No. L 259/10-11.

Commission Decision 2008/798/EC imposing special conditions governing theimport of products containing milk or milk products originating in or consignedfrom China, and repealing Commission Decision 2008/757/EC (2008b). OfficialJournal of the European Union, No. L 273/18-20.

Commission Decision 2008/921/EC amending Decision 2008/798/EC (2008c).Official Journal of the European Union, No. L 331/19-20.

Ehling, S., Tefera, S., & Ho, I. P. (2007). High-performance liquid chromatographicmethod for the simultaneous detection of the adulteration of cereal flours withmelamine and related triazine by-products ammeline, ammelide, and cyanuricacid. Food Additives and Contaminants, 24, 1319–1325.

Fodey, T. L., Thompson, C. S., Traynor, I. M., Haughey, S. A., Kennedy, D. G., & Crooks,S. R. H. (2011). Development of an optical biosensor based immunoassay to

screen infant formula milk samples for adulteration with melamine. AnalyticalChemistry, 83(12), 5012–5016.

Garber, E. A. (2008). Detection of melamine using commercial enzyme-linkedimmunosorbent assay technology. Journal of Food Protection, 71, 590–594.

Garber, E. A. E., & Brewer, V. A. (2010). Enzyme-linked immunosorbent assaydetection of melamine in infant formula and wheat food products. Journal ofFood Protection, 73(4), 701–707.

Gossner, C. M.-E., Schlundt, J., Embarek, P. B., Hird, S., Lo-Fo-Wong, D., Betran, J. J. O.,et al. (2009). The melamine incident: Implications for international food andfeed safety. Environmental Health Perspectives, 117(12), 1803–1808.

Lei, H., Shena, Y., Songa, L., Yanga, J., Chevallier, O. P., Haughey, S. A., et al. (2010).Hapten synthesis and antibody production for the development of a melamineimmunoassay. Analytica Chimica Acta, 665, 84–90.

Lei, H., Su, R., Haughey, S. A., Wang, Q., Yang, J., Xu, Z., et al. (2011). Development of aspecifically enhanced enzyme-linked immunosorbent assay for the detection ofmelamine in milk. Molecules, 16, 5591–5603.

Lin, M. (2009). A review of traditional and novel detection techniques for melamineand its analogues in foods and animal feed. Frontiers of Chemical Engineering inChina, 3(4), 427–435.

Mauer, L. J., Chernyshova, A. A., Hiatt, A., Deering, A., & Davis, R. (2009). Melaminedetection in infant formula powder using near- and mid-infrared spectroscopy.Journal of Agricultural and Food Chemistry, 57, 3974–3980.

Miao, H., Fan, S., Wu, Y.-N., Zhang, L., Zhou, P.-P., Li, J. G., et al. (2009). Simultaneousdetermination of melamine, ammelide, ammeline, and cyanuric acid in milkand milk products by gas chromatography–tandem mass spectrometry.Biomedical and Environmental Sciences, 22, 87–94.

Owens, B., McCann, M. E. E., McCracken, K. J., & Park, R. S. (2009). Prediction ofwheat chemical and physical characteristics and nutritive value by nearinfrared reflectance spectroscopy. British Poultry Science, 50(1), 103–122.

Pei, X., Tandon, A., Alldrick, A., Giorgi, L., Huang, W., & Yang, R. (2011). The Chinamelamine milk scandal and its implications for food safety regulation. FoodPolicy, 36, 412–420.

Sharma, K., & Paradakar, M. (2010). The melamine adulteration scandal. FoodSecurity, 2, 97–107.

Sierra, V., Aldai, N., Castro, P., Osoro, K., Coto-Montes, A., & Olivan, M. (2008).Prediction of fatty acid composition of beef by near infrared transmittancespectroscopy. Meat Science, 78(3), 248–255.

Vaclavik, L., Rosmus, J., Popping, B., & Hajslova, J. (2010). Rapid determination ofmelamine and cyanuric acid in milk powder using direct analysis in real time-time-of-flight mass spectrometry. Journal of Chromatography, A, 1217(25),4204–4211.

Venkatasami, G., & Sowa, J. R. Jr., (2010). A rapid, acetonitrile-free, HPLC method fordetermination of melamine in infant formula. Analytica Chimica Acta, 665(2),227–230.

Wang, Q., Haughey, S. A., Sun, Y., Eremin, S. A., Li, Z., Liu, H., et al. (2011).Development of a fluorescence polarization immunoassay for the detection ofmelamine in milk and milk powder. Analytical and Bioanalytical Chemistry, 399,2275–2284.

Yan, N., Zhou, L., Zhu, Z., & Chen, X. (2009). Determination of melamine in dairyproducts, fish feed, and fish by capillary zone electrophoresis with diode arraydetection. Journal of Agricultural and Food Chemistry, 57, 807–811.

Yin, W. W., Liu, J. T., Zhang, T. C., Li, W. H., Liu, W., Meng, M., et al. (2010).Preparation of monoclonal antibody for melamine and development of anindirect competitive ELISA for melamine detection in raw milk, milk powder,and animal feeds. Journal of Agricultural and Food Chemistry, 58(14), 8152–8157.