Rice Science, 2015, 22(5): 245249

Copyright 2015, China National Rice Research Institute

Hosting by Elsevier B.V. All rights reserved
DOI: 10.1016/S1672-6308(14)60291-2

Discrimination of Transgenic Rice Based on Near Infrared

Reflectance Spectroscopy and Partial Least Squares
Regression Discriminant Analysis

ZHANG Long1, 3, WANG Shan-shan2, DING Yan-fei2, PAN Jia-rong2, ZHU Cheng2, 3
(1College of Ecology, Lishui University, Lishui 323000, China; 2College of Life Sciences, China Jiliang University, Hangzhou
310018, China; 3College of Life Sciences, Zhejiang University, Hangzhou 310058, China)

Abstract: Near infrared reflectance spectroscopy (NIRS), a non-destructive measurement technique, was
combined with partial least squares regression discrimiant analysis (PLS-DA) to discriminate the transgenic
(TCTP and mi166) and wild type (Zhonghua 11) rice. Furthermore, rice lines transformed with protein gene
(OsTCTP) and regulation gene (Osmi166) were also discriminated by the NIRS method. The performances of
-1 -1
PLS-DA in spectral ranges of 4 0008 000 cm and 4 00010 000 cm were compared to obtain the optimal
spectral range. As a result, the transgenic and wild type rice were distinguished from each other in the range of
-1
4 00010 000 cm , and the correct classification rate was 100.0% in the validation test. The transgenic rice
-1
TCTP and mi166 were also distinguished from each other in the range of 4 00010 000 cm , and the correct
classification rate was also 100.0%. In conclusion, NIRS combined with PLS-DA can be used for the
discrimination of transgenic rice.
Key words: near infrared reflectance spectroscopy; genetically-modified food; regulation gene; protein gene;
partial least squares regression discrimiant analysis

Genetically-modified foods (GMOs) are inserted with foreign
genes which increase resistance to diseases, pests and
herbicides or improve nutritional contents which never occur
naturally (Anklam et al, 2002). Since it is suspected with the
safety about environmental hazards, human health risks, and
economic concerns, the public are scared of consuming GMOs,
and the GMOs are severely rejected in many regions of the
world. In order to regulate the introduction and production of
GMOs, many regulations and legislations are promulgated with
the Ministry of Health, the Ministry of Agriculture, the General
Administration of Quality Supervision, and the Inspection and
Quarantine in China. Thus, the fast reliable detection methods
must be developed to support the regulations and legislations
above.
There are many detection methods for transgenic food.
Western blot (Lipton et al, 2000; Sambrook and Russel, 2000),
enzyme linked immunosorbent assays (Yates, 1999; Margarit
et al, 2006) and lateral flow strip (Fagan et al, 2001) are protein-
based methods. Southern blot (Ross et al, 1999; Stull, 2001),
qualitative polymerase chain reaction (qPCR) (van Hoef et al,
1998; Lipp et al, 1999; Singh et al, 2007), microarray (Miraglia
et al, 2004) and real time PCR (Heid et al, 1996; Ahmed, 2000;
Mde et al, 2006; Akiyama et al, 2007; Grohmann and Mde,
2009) are DNA-based methods. Additionally, high performance
Received: 11 November 2014; Accepted: 26 May 2015
Corresponding author: ZHU Cheng (pzhch@zju.edu.cn)
liquid chromatography and gas chromatography are also useful
technologies (Alishahi et al, 2010). However, these technologies
had many shortcomings, such as high cost, difficult to use,
special need, and long duration.
Near infrared reflectance spectroscopy (NIRS) is sensitive to
organic compounds with vibration overtones of CH, OH and
NH, which is rapid, lower cost and non-destructive. The
content of modified DNA is in amount of ultra trace, so
detection of the modified DNA is difficult to identify the
GMOs, while detection of the products resulting from the
genetic modification (particular proteins) or larger structural
changes (phenotype) are feasible (Munck et al, 2001; van Duijn
et al, 2002; Xie et al, 2007a, b; Alishahi et al, 2010). In
previous studies, NIRS has successfully used in detection of
transgenic tomato (Xie et al, 2007a), barley (Hurburgh et al,
2000) and corn (Rossel et al, 2001) or mutant of barley (Munck
et al, 2004) and corn (Campbell et al, 2000).
Rice is relatively easy to consult transgenic. There are rarely
studies conducted on the detection of transgenic rice to our
knowledge (Jiao et al, 2010). Translationally-controlled tumor
protein (TCTP) is most likely controlled by a cell housekeeping
gene which presents as a multifunctional protein and essential
in the development of mammals, higher plants and
Saccharomyces cerevisiae. It was recently discovered that
OsTCTP works on mercury resistance. As we all known,
microRNAs (miRNAs) are a class of small non-coding RNAs
which negatively regulate specific target mRNAs at the
246
post-transcriptional level. Thus, cd-responsive gene Osmi166
were selected (Ding et al, 2011).
In the present study, heavy metal responsive genes OsTCTP
and Osmi166 transgenic rice were employed to discriminate the
transgenic rice (Wang, 2010; Ding et al, 2011). The aim of the
present study was to discriminate the transgenic and wild type
rice, and the transgenic rice transformed with protein gene and
regulation gene. Additionally, the partial least squares
regression discrimiant analysis (PLS-DA) was employed as a
discriminant model.
MATERIALS AND METHODS
Transgenic rice
Wild type rice Zhonghua 11 (Oryza sativa L. subsp. japonica,
ZH11) was used as the transgenic material. Two single-copy
transgenic rice lines were developed in the authors laboratory
by respectively introducing OsTCTP and Osmi166 genes into
ZH11 using an Agrobacterium-mediated transformation method.
The rice seed used were all T3 homozygous lines and
designated as TCTP and mi166, respectively. OsTCTP and
Osmi166 genes expressed in rice plants and the expression
levels were stable. The rice lines were all planted and harvested
under natural condition and dried in the sun in the same field.
NIRS analysis
The near infrared reflectance spectra of rice grains were
scanned on a Nicolet Nexus 870 FT-IR spectrometer (Thermo
Corporation, USA) with the mode of diffuse transmission. The
spectra of 192 rice grain samples were measured at 8 cm-1
resolution with 32 scans above the detection window. And both
the obverse and the reverse of the grains (three grains for each
sample) were scanned. The spectra were collected with the
OMNIC 6.0 software in range of 4 00010 000 cm-1 in the
absorbance mode. Each spectrum was structured in 3040 data
with the format of ASCII which are easily combined with the
Matlab software version 6.5.
PLS-DA
PLS is a well-established multivariate regression model which
constructs a mathematic relationship between descriptors and
dependent variables (Kleinbaum et al, 1988). In the PLS model,
principal component analysis (PCA) was first processed, and
then the scores of principal components (PCs) obtained were
considered as new eigenvectors of the original spectra. The
performance of PLS model is affected by the number of PCs
(Marengo et al, 2008). In order to discriminate the transgenic
and wild type rice, PLS was used as discrimination by designed
the values of different category. The values of transgenic
(TCTP and mi166) and wild type (ZH11) rice were designated
as -1 and 1, respectively, in the PLS-DA model. To identify
rice transformed with protein gene and regulation gene, TCTP
and mi166 were used and the values were designated as -1
Rice Science, Vol. 22, No. 5, 2015
and 1 in the PLS-DA model, respectively. The predicted
values below 0 were considered to the category of -1, and
the values above 0 were considered to the category of 1.
Leave-one-out cross-validation was used for the calibration,
which involved in using a single observation from the original
sample as the validation data and the remaining observations as
the training data.
Model evaluation
The PLS-DA model was evaluated with the parameters of
root-mean-square error (RMSE) and correlation coefficient (r2).
RMSE used for calibration was designated as root mean square
error in calibration (RMSEC) and for validation or prediction
as root mean square error in prediction (RMSEV). r2 used for
calibration was designated as r2c and for validation as r2v.
RESULTS AND DISCUSSION
Diffuse transmittance spectra of rice grains
Spectra of transgenic rice and wild types ZH11 in the region of
4 00010 000 cm-1 are shown in Fig. 1. It is obviously that the
shapes of the spectra of TCTP, mi166 and ZH11 were
overlapped and quite homogeneous and cannot be identified by
naked eyes, but there were some variants in the 4 00010 000
cm-1 region. The noise of spectra in the 8 00010 000 cm-1
region was a little more than that in the 4 0008 000 cm-1
region. Therefore, in the following calculation, the performance
of the model was compared between the 4 0008 000 cm-1 and
4 00010 000 cm-1 regions.
PCA of rice grain spectra
To examine the spectral variability of the selected samples,
PCA was performed on the raw spectra across the spectral
range 4 00010 000 cm-1. The first two principal components
explained 62.90% of the variations in the raw spectra of rice
grain samples (PC1, 54.51%; PC2, 8.39%) (Fig. 2). The score
plot of the first two PCs showed that the TCTP, mi166 and
2
1
Absorbance
0
-1
-2
-3
4 000 6 000 8 000 10 000
Wavenumber (cm-1)
1 grains.
Fig. 1. Standard normal variance pretreated spectra of rice
ZH11, Zhonghua 11.
ZHANG Long, et al. Near Infrared Reflectance Spectroscopy for Discrimination of Transgenic Rice Seeds 247

ZH11 sets were overlapped in the PC1 and PC2 spaces (Fig. 3),
and the three kinds of rice cannot be discriminated from each 80
other clearly.

Cumulative variance (%)

60
Distinction between transgenic and wild type and between
TCTP and mi166 with PLS-DA model 40

The number of PCs in the PLS-DA model is very important 20

because very few components will generate an under-fitted
model, i.e., fits loosely the data structure. However, using too 0
many components generates an over-fitted model, one which
fits parts of the noise of the calibration set, thus generating a 0 2 4 6 8 10

low RMSEC but performing poorly in the validation set (Broomhead Number of PCs

and Lowe, 1988). Different ranges of spectra contained the Fig. 2. Variation explained by principal component1 (PC) in
variant information of rice grains. In the present study, two principal component analysis.
regions of spectra 4 0008 000 cm-1 and 4 00010 000 cm-1
were compared to evaluate the optimal range spectra, which
could get the best discrimination result. In the experiment of mi166
discriminate, for transgenic and wild type rice, in the range of TCTP
Zhonghua 11
4 0008000 cm-1, the optimal numbers of PC were five in the
both calibration and validation tests, whereas in the range of

Score of PC2
4 00010 000 cm-1, the optimal numbers of PC were six in the
both calibration and validation tests (Table 1). In the experiment
of discriminating rice transformed with protein (TCTP) or
regulation genes (mi166), the optimal numbers of PC were all
five in the both calibration and validation tests for the two
wavenumber ranges (Table 2).
PLS-DA was constructed to distinguish rice from those with -7 0 7
transformed foreign genes. The performance of PLS-DA was Score of PC1
evaluated by the values of RMSE and r2 in the calibration and
1
Fig. 3. Plots of the first and second principal components (PCs).
validation tests. The spectral range which got the lowest values
of RMSE and the highest values of r2 showed the best
performance of model. It was obviously that spectra in the value of RMSEV and higher value of r2v, which were 0.2878
range of 4 00010 000 cm-1 showed lower value of RMSEC and 0.8979, respectively. Thus, the optimal bands responsible
and higher value of r2c, which were 0.2695 and 0.9183, for the discrimination of transgenic and wild type rice were in
respectively, in the discrimination of transgenic and wild type the range of 4 00010 000 cm-1. The correct classification of
rice. Similarly, the result of validation test showed the lower calibration test and validation test all achieved 100% perfectly

Table 1. Effects of different wavenumber range on classification results of partial least squares regression discrimiant analysis models
between transgenic and wild type rice.
Wavenumber Calibration Validation
(cm-1) No. of PCs RMSEC r2
c CRc (%) No. of PCs RMSEV r2v CRv (%)
4 0008 000 5 0.3555 0.8578 100 5 0.3543 0.8344 100
4 00010 000 6 0.2695 0.9183 100 6 0.2878 0.8979 100
PC, Principal component; RMSEC, Root mean square error in calibration; r2c, Correlation coefficient of calibration; CRc, Correct rate of
calibration; RMSEV, Root mean square error in validation; r2p, Correlation coefficient of validation; CRv, Correct rate of validation.

Table 2. Effects of different wavenumber range on classification results of partial least squares regression discrimiant analysis models
between TCTP and mi166.
Wavenumber Calibration Validation
(cm-1) No. of PCs RMSEC r2c CRc (%) No. of PCs RMSEV r2v CRv (%)
4 0008 000 5 0.3659 0.8661 97.5 5 0.3799 0.8123 98.3
4 00010 000 5 0.2945 0.9133 97.5 5 0.3873 0.7537 100.0
PC, Principal component; RMSEC, Root mean square error in calibration; r2c, Correlation coefficient of calibration; CRc, Correct rate of calibration;
RMSEV, Root mean square error in validation; r2v, Correlation coefficient of validation; CRv, Correct rate of validation.
248
1 rate of validation test all achieved 100.0% in the spectral range
Predicted value
0 identification of transgene and wild type rice as well as rice
-1
-1
Designed value
Fig. 4. Regression plot of designed and prediction 1category
variables of transgenic (TCTP and mi166) and wild type
(Zhonghua 11) rice.
1
Predicted value
0 Environ Sci Health Part C, 18(2): 75125.
-1
-1 1
Designed value
Fig. 5. Regression plot of reference and prediction1 category
variables of TCTP and mi166.
(Table 1).
In order to discriminate TCTP and mi166 which were
products of two kinds of transplants in the seeds, the spectra of
TCTP and mi166 were calculated. Table 2 showed that the
spectra in the range of 4 00010 000 cm-1 had lower value of
RMSEC and higher value of r2c, which were 0.2945 and 0.9133
in the calibration test, respectively, by the way, in the validation
test, lower value of RMSEV and higher value of r2v were
shown in the range of 4 0008 000 cm-1, while the correct
classification rate in the range of 4 00010 000 cm-1 was 100.0%,
which was higher than that in the range of 4 0008 000 cm-1.
Therefore, It was concluded that the optimal bands responsible
for the discrimination of TCTP and mi166 were in the range of
4 00010 000 cm-1. The correct classification of calibration test
and validation test were 97.5% and 100.0%, respectively (Table
2). The plots of reference values vs. predicted values for
PLS-DA in the ranges of 4 00010 000 cm-1 are showed in Figs.
4 and 5, respectively.
CONCLUSIONS
NIRS coupled with PLS-DA can be used to distinguish the
Rice Science, Vol. 22, No. 5, 2015
transgenic rice TCTP and mi166 from the wild type rice, and
also to differentiate TCTP from mi166. The correct classification
of 4 00010 000 cm-1. Thus, it might be possible to apply the
non-destructive technology NIRS and PLS-DA model in the
transformed with different kinds of genes.
ACKNOWLEDGEMENTS
1
The research was supported by the projects under the
Innovation Team of the Safety Standards and Testing Technology
for Agricultural Products of Zhejiang Province, China (Grant
No. 2010R50028), and the National Key Technologies R&D
Program of China during the 11th Five-Year Plan Period (Grant
No. 2006BAK02A18).
REFERENCES
Ahmed F E. 2000. Molecular markers for early cancer detection. J
Akiyama H, Sasaki N, Sakata K, Ohmori K, Toyota A, Kikuchi Y,
Watanabe T, Furui S, Kitta K, Maitani T. 2007. Indicated
detection of two unapproved transgenic rice lines contaminating
vermicelli products. J Agric Food Chem, 55(15): 59425947.
Alishahi A, Farahmand H, Prieto N, Cozzolino D. 2010.
Identification of transgenic foods using NIR spectroscopy: A
review. Spectrochimica Acta Part A, 75: 17.
Anklam E, Gadani F, Heinze P, Pijnenburg H, van den Eede G.
2002. Analytical methods for detection and determination of
genetically modified organisms in agricultural crops and
plant-derived food products. Eur Food Res Technol, 214: 326.
Broomhead D S, Lowe D. 1988. Multi-variable functional
interpolation and adaptive networks. Comp Syst, 2(3): 269303.
Campbell M R, Sykes J, Glover D V. 2000. Classification of single
and double-mutant corn endosperm genotypes by near-infrared
transmittance spectroscopy. Cereal Chem, 77: 774778.
Ding Y F, Chen Z, Zhu C. 2011. Microarray-based analysis of
cadmium-responsive microRNAs in rice (Oryza sativa). J Exp
Bot, 62(10): 35633573.
Fagan J, Schoel B, Haegert A, Moore J, Beeby J. 2001. Performance
assessment under field conditions of a rapid immunological test
for transgenic soybeans. Int J Food Sci Technol, 36: 111.
Grohmann L, Mde D. 2009. Detection of genetically modified
rice: Collaborative validation study of a construct-specific
real-time PCR method for detection of transgenic Bt rice. Eur
Food Res Technol, 228(3): 497500.
Heid C A, Stevens J, Livak K J, Williams P M. 1996. Real-time
quantitative PCR. Genome Res, 6: 986994.
Hurburgh C R, Rippke G R, Heithoff C, Roussel S A, Hardy C L.
2000. Detection of genetically modified grains by near-infrared
spectroscopy. In: Proceedings PITTCON 2000: Science for the
21st Century. 1217 March, New Orleans, USA: 1431.
ZHANG Long, et al. Near Infrared Reflectance Spectroscopy for Discrimination of Transgenic Rice Seeds
Jiao Z, Si X X, Li G K, Zhang Z M, Xu X P. 2010. Unintended
compositional changes in transgenic rice seeds (Oryza sativa L.)
studied by spectral and chromatographic analysis coupled with
chemometrics methods. J Agric Food Chem, 58(3): 17461754.
Kleinbaum D, Kupper L, Muller K 1988. Applied Regression
Analysis and Other Multivariate Methods. 2nd edn. PWS-Kent,
Boston: 657662.
Lipp M, Brodmann P, Pietsch K, Pauwels J, Anklam E, Brchers T,
Braunschweiger G, Busch U, Eklund E, Eriksen F D. 1999.
IUPAC collaborative trial study of a method to detect genetically
modified soybeans and maize in dried powder. J AOAC Int, 82:
923928.
Lipton C R, Dautlick J X, Grothaus G D, Hunst P L, Magin K M,
Mihaliak C A, Rubio F M, Stave J W. 2000. Guidelines for the
validation and use of immunoassays for determining of
introduced proteins in biotechnology enhanced crops and
derived food ingredients. Food Agric Immunol, 12: 153164.
Mde D, Degner C, Grohmann L. 2006. Detection of genetically
modified rice: A construct-specific real-time PCR method based
on DNA sequences from transgenic Bt rice. Eur Food Res
Technol, 224(2): 271278.
Marengo E, Robotti E, Bobba M, Milli A, Campostrini N, Righetti
S C, Cecconi D, Righetti P G. 2008. Application of partial least
squares discriminant analysis and variable selection procedures:
A 2D-PAGE proteomic study. Anal Bioanal Chem, 390:
13271342.
Margarit E, Reggiardo M I, Vallejos R H, Permingeat H R. 2006.
Detection of BT transgenic maize in foodstuffs. Food Res Int, 39:
250255.
Miraglia M, Berdal K G, Brera C, Corbisier P, Holst-Jensen A, Kok
E J, Marvin H J, Schimmel H, Rentsch J, van Rie J P, Zagon J.
2004. Detection and traceability of genetically modified
organisms in the food production chain. Food Chem Toxicol, 42:
11571180.
Munck L, Pram Nielsen J, Mller B, Jacobsen S, Sndergaard I,
Engelsen S B, Nrgaard L, Bro R. 2001. Exploring the
phenotypic expression of a regulatory proteome-altering gene by
spectroscopy and chemometrics. Anal Chim Acta, 446: 171186.
Munck L, Mller B, Jacobsen S, Sndergaard S. 2004. Near
infrared spectra indicate specific mutant endosperm genes and
249
reveal a new mechanism for substituting starch with (1/3,
1/4)--glucan in barley. J Cereal Sci, 40: 213222.
Ross R, Ross X L, Rueger B, Laengin T, Reske-Kunz A B. 1999.
Nonradioactive detection of differentially expressed genes using
complex RNA or DNA hybridization probes. Biotechnology, 26:
150155.
Rossel S A, Hardy C L, Hurburgh C R, Rippke G R. 2001.
Application of near-infrared diffuse reflectance spectroscopy to
the detection and identification of transgenic corn. Appl Spectr,
55: 14251432.
Sambrook J, Russel D. 2000. Molecular Cloning: A Laboratory
Manual. 3rd edn. New York: Cold Spring Harbor Laboratory
Press.
Singh C K, Ojha A, Kachru D N. 2007. Detection and
characterization of cry1Ac transgene construct in Bt cotton:
Multiple polymerase chain reaction approach. J AOAC Int, 90(6):
15171725.
Stull D. 2001. A feat of fluorescence. Scientist, 15: 2021.
van Duijn G J, van Biert R, Bleeker-Marcelis H, van Boeijen I,
Adan A J, Jhakrie S, Hessing M. 2002. Detection of genetically
modified organisms in foods by protein- and DNA-based
techniques: Bridging the methods. J AOAC Int, 85(3): 787791.
van Hoef A M A, Kok E J, Bouw E, Kuiper H A, Keijer J. 1998.
Development and application of a selective detection method for
genetically modified soy and soy-derived products. Food Addit
Contam, 15(7): 767774.
Wang F J. 2010. Comparative Proteomic Study and Functional
Analysis of Tolerance-Related Genes from Rice Roots During
Hg2+ Stress. Hangzhou: Zhejiang University. (in Chinese with
English abstract)
Xie L J, Ying Y B, Ying T J. 2007a. Combination and comparison
of chemometrics methods for identification of transgenic
tomatoes using visible and near-infrared diffuse transmittance
technique. J Food Engin, 82: 395401.
Xie L J, Ying Y B, Ying T J, Yu H Y, Fu X P. 2007b. Distrimination
of transgenic tomatoes based on visible/near-infrared spectra.
Anal Chim Acta, 584(2): 379384.
Yates K. 1999. Detection Methods for Novel Foods Derived from
Genetically Modified Organisms. ILSI, Europe: 339349.