- Email: [email protected]
Nondestructive assessment of amino acid composition in rapeseed meal based on intact seeds by near-infrared reflectance spectroscopy
Animal Feed Science and Technology 165 (2011) 111–119
Contents lists available at ScienceDirect
Animal Feed Science and Technology journal homepage: www.elsevier.com/locate/anifeedsci
Nondestructive assessment of amino acid composition in rapeseed meal based on intact seeds by near-infrared reflectance spectroscopy G.L. Chen a,b , B. Zhang a , J.G. Wu a,∗ , C.H. Shi a,∗ a b
Department of Agronomy, Zhejiang University, Hangzhou 310029, China School of Agriculture and Food Science, Zhejiang Forestry University, Linan 311300, China
a r t i c l e
i n f o
Article history: Received 20 January 2010 Received in revised form 25 January 2011 Accepted 4 February 2011
Keywords: Near infrared reflectance spectroscopy (NIRS) Amino acids content Rapeseed meal Intact rapeseed
a b s t r a c t The ability of near-infrared spectroscopy (NIRS) was tested for estimating individual and total amino acid contents in rapeseed meal. Twelve different amino acids (aspartic acid, threonine, serine, glutamic acid, glycine, isoleucine, leucine, tyrosine, phenylalanine, lysine, histidine and arginine) in rapeseed meal could be predicted. The R2 ranged from 0.89 to 0.98, 1 − VR (1 minus the ratio of unexplained variance to total variance) ranged from 0.86 to 0.97 and the ratio of sample standard deviation (SD) to the standard error of cross-validation (SECV) ranged from 2.69 to 5.90. The equations for alanine showed better agreement between reference value and spectra (R2 = 0.84, 1 − VR = 0.82 and SD/SECV = 2.15). The calibration model for proline (R2 = 0.81 and 1 − VR = 0.78), methionine (R2 = 0.80 and 1 − VR = 0.78), cysteine (R2 = 0.76 and 1 − VR = 0.74), and valine (R2 = 0.62 and 1 − VR = 0.58) however explained less variation. The NIRS prediction equation for total amino acid also showed high coefficient of determination (R2 = 0.93) and SD/SECV (3.87), and low SECV (17.01 g/kg). Equations of 9 amino acids (aspartic acid, glutamic acid, glycine, alanine, valine, leucine, lysine, histidine and arginine) were developed for relative contents of total amino acid and deemed useful for prediction with R2 values from 0.80 to 0.95, 1 − VR from 0.70 to 0.95 and SD/SECV from 1.83 to 3.95 and reasonably low SECV values. These results demonstrated that NIRS is a reliable tool for nondestructive assessment of variation in amino acid contents, increasing the efficiency of breeding and accelerating the selection process in rapeseed. © 2011 Elsevier B.V. All rights reserved.
1. Introduction Brassica oilseed crops are mainly grown for the production of vegetable oils which are used for human consumption or industrial applications and their byproduct are high protein feed meal which includes about 400 g/kg of crude protein. Rapeseed protein which has a rational amino acid composition is another important source of nutrition (Goding et al., 1972; Huisman and Tolman, 1990). The contents of two amino acids (methionine and cysteine) are higher than those in soybean and peanut. Soluble protein, lysine and other essential amino acid contents are also higher than those in sunflower and
Abbreviations: 1 − VR, 1 minus the ratio of unexplained variance to total variance; CV, the coefficient of variation; MPLS, modified partial least squares; NIRS, near infrared reflectance spectroscopy; R2 , coefficient of determination in calibration; RSDr , relative standard deviation for repeatability; SD, standard deviation; SD/SECV, ratio of the standard deviation (SD) of the amino acid contents in the calibration samples to the standard error of cross-validation (SECV); SEC, the standard error of calibration; SECV, standard error of cross-validation; SNV + D, standard normal variate + detrending; Sr , repeatability of standard deviation; TAA, total of amino acids. ∗ Corresponding authors. Tel.: +86 571 86971691; fax: +86 571 86971117. E-mail addresses: [email protected] (J.G. Wu), [email protected] (C.H. Shi). 0377-8401/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.anifeedsci.2011.02.004
112
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
sesame. The nutritional value of rapeseed meal is not less than that of soybean (Josefesson and Muhlenberg, 1968; Simbaya et al., 1995). Wu et al. (2005) found that protein content of rapeseed was simultaneously controlled by genetic effects of embryo, cytoplasm, and maternal plant, of which the maternal genetic effect was most important. The total narrow-sense heritability is high. Röbbelen (1981) pointed out that the development of high-protein cultivars in rapeseed is an important objective for breeding because the rapeseed meal can be transformed into good feed for livestock. High protein content and rational amino acids composition in rapeseed are major concern to the breeder. Improvement for seed quality traits depends on the availability of fast and accurate methods for their measurement in breeding practice (Wu and Shi, 2004). Near infrared reflectance spectroscopy (NIRS) technique started in 1960s (Ben-Gera and Norris, 1968), has developed rapidly since late 1980s, and become a powerful green analytical tool for the routine analysis of quality traits in rapeseed. Using intact seeds sample, NIRS calibration equations have been confirmed to be useful to routinely determine simultaneously several quality traits, such as oil, protein and glucosinolate contents as well as fatty acid composition (Velasco et al., 1997; Velasco and Becker, 1998a,b; Velasco et al., 1999; Wu et al., 2002a). Velasco and Möllers (2002) reported that the NIRS technique was reliable enough for the estimation of protein content in a single rapeseed. Hom et al. (2007) compared the accuracy of NIRS analysis of protein and oil content in intact seeds sample size of 3–4 g with the same content in a single seed (5 mg), and concluded that standard NIRS calibration equations could be used for screening a single rapeseed. The method to determine the fatty acid composition of the seed oil at a half-seed level (Downey and Harvey, 1963) has been substituted by the above method for the improvement of seed oil quality. Near-infrared spectroscopy has been used on intact grains of wheat (Abe et al., 1995; Roussel et al., 2005), milled rice flour (Wu et al., 2002b) and a single corn kernel (Baye et al., 2006) to predict nutrient quality traits such as protein and amino acid content. In the feed industry, quick and accurate analysis methods on contents of essential amino acids in the most important protein-rich feed draws a lot of attention from feed researchers. NIRS calibrations were developed for the accurate and fast determination of the total contents of important essential amino acids and protein with the finely ground material in the most cereals, bran and meals for animal feed production (Fontaine et al., 2001, 2002). Their equations facilitate the routine work, improve the accuracy of feed formulation and quality, and decrease production cost (Pujol et al., 2007). NIRS technique is widely applied in agricultural product analyses and breeding programs (Wu et al., 2002a,b). Protein content is determined in the laboratory by a classical procedure of the Kjeldahl method, and amino acid content by the HCl hydrolysis-HPLC method. These conventional chemical analytical methods are time-consuming and expensive. They are also considered destructive because they require grinding and other pretreatments of samples which make them unsuitable for the analysis of a large number of samples in the early generations of breeding. A rapid and accurate method to identify and screen breeding materials without sample preparation is crucial for the efficiency in breeding and an acceleration of the selection process. Considering the demands of breeding a rapeseed cultivar with high quality protein and its genetic analysis, it is necessary to develop a fast and efficient method such as near-infrared reflectance spectroscopy technology to determine amino acids composition in rapeseed meal. To date, no attempt has been made to determine amino acids content in rapeseed meal based on intact seeds by NIRS. It is for this reason that we have focused on the technology of NIRS. This work was to study the potentials of NIRS to estimate amino acid contents of rapeseed meal and to explore its applicability in identifying variability for these traits. 2. Materials and methods 2.1. Materials An original population of 621 rapeseed samples was mainly derived from genetic experiments and the breeding programs. All seed samples including F1 and F2 hybrid generations and their parents were collected in the years of 2007 and 2008. Being cultivars and breeding line across environments the seed samples had a larger variation in seed qualities making them suitable for NIRS calibration. All the raw samples were scanned to collect the NIRS spectra. 2.2. Collection of spectra WinISI II (Version1.04) software was applied to collect spectra and develop the calibration equations in this study. Whole intact rapeseed samples were scanned on a NIRSystems model 5000 monochromator (NIRSystems Inc., Silver Spring, MD, USA). Approximately 3 g of intact rapeseed samples were placed in a small ring cup of 36 mm inner diameter, and reflectance spectra (log 1/R) from 1100 to 2498 nm were recorded at 2 nm wavelength increment. Each sample was subsequently scanned 32 times in small ring cells (Shenk and Westerhaus, 1993). 2.3. Selection of calibration set A powerful approach to choose samples based on their spectra has been developed with the CENTER and SELECT algorithm by Shenk and Westerhaus (1991a). After the original population with 621 samples was scanned, the algorithm CENTER was used for the calculation of principal components and GH for the description of spectral boundaries and detection of outliers (Shenk and Westerhaus, 1991a). The number of samples reduced to 611 after outliers were excluded. Following centering of
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
113
the data, the SELECT algorithm was used for grouping the spectra into clusters with NH of 0.8. One representative spectrum from each cluster was selected for the arrangement of the calibration set of samples. The number of collection decreased to 238. After 12 samples were eliminated for possessing insufficient amount for determination, 226 samples were analyzed for amino acid composition by the reference method. Therefore, based on the spectra variation, 226 samples were selected for the development of a calibration equation for amino acid content in this study. 2.4. Determination of amino acids for rapeseed meal To develop rapeseed meals for each rapeseed sample, whole intact rapeseed samples were ground through a 1 mm screen and oil was extracted by Soxhlet method (AOAC, 1980) with crude fat analyzer model SZF-06 (Shanhai Xinjia Electronic Co. Ltd, China).About 3 g milled rapeseeds sample wrapped were submerged in 50 ml ether solvent for 3 h at 75 ◦ C. After oil removal, the remains were dried to constant weighed for 1 h at 105 ◦ C to be suitable for estimating amino acids content. About 30 mg of each sample meal was weighed with a readability of 0.01 mg and 10 ml of 6 M HCl was added. The air pressure was reduced to 7 Pa with vacuum pump. The test tube was enclosed by alcohol lamp with high temperature and placed into constant temperature box for hydrolysis 24 h at 105 ◦ C. Sample liquid was cooled, mixed, and diluted to 100 ml. Ten milliliter dilution was transferred to distillation flask which was held to a rotary evaporator (model RE-52AA, Shanghai, China) for evaporation with two replications. The acid was evaporated in water bath at 65 ◦ C and the residue was dissolved in 10 ml 0.02 M HCL buffer. The sample liquid was mixed by ultrasonic instrument and filtered with 0.45 m film. About 1 ml of each sample was injected into an auto-sampler bottle for analysis. The contents of 17 amino acids were measured for each sample using amino acid auto-analyzer model L-8900 (Hitachi High-Technologies Corporation, Japan). Above method was similar to that for determination of amino acids in milled rice (Wu et al., 2002a,b). The amount of each amino acid in the samples was calculated with reference to the standard solution (Type H, Wako pure chemical Industrial Corporation, Japan) and expressed as g/kg dried meal weight (DW). It was called absolute content for each amino acid. Total amino acid (TAA) was the sum of 17 amino acids in each sample meal. The relative content of each amino acid was expressed as g/kg TAA in each sample. All values were given on basis of a dry meal. The standard solution of each amino acid was 0.1 M. The Sr , the repeatability of standard deviation, defined as the mean of the absolute difference d2 /2N, where d is difference between between laboratory duplicates, was calculated using the following formula: Sr = two duplicates on an ith sample, and N is the number of samples. The relative standard deviation (RSDr = (Sr /mean) × 1000) for repeatability with unit of g/kg (Theander et al., 1995). 2.5. Data processing Calibration models for amino acids were developed using the platform of WinISI II (Version1.04) software. The spectral data with range from1100 to 2498 nm and an interval of 2 nm were subjected to two different mathematical treatments during the development of calibration models. For example in the notation (1,4,4,1), the first number indicates the order of the derivative of log(1/R), the second is the gap in data points over which the derivative is calculated, the third and fourth numbers refer to the number of data points used in the first and second smoothing, respectively. In addition, standard normal variate and detrending (SNV + D) were applied to correct scattering (Barnes et al., 1989). The SNV + D were designed to remove additive baseline and multiplicative signal effects resulting in a spectrum with zero mean and a variance equal to 1. The application of SNV + D to raw spectral data reduces the differences in spectra related to physical characteristics such as particle size. A modified partial least squares algorithm was used (Shenk and Westerhaus, 1991b). Calibrations were tested using cross-validation. Two rounds of outlier elimination were allowed, with samples with an ‘H’ value larger than 4 (spectral outliers) and a (Student) ‘t’ value larger than 2.5 (sample which did not fit the calibration model) being eliminated. The standard error of calibration (SEC), coefficient of determination (R2 ), standard error of cross-validation (SECV) and 1 − VR (1 minus the ratio of unexplained variance to total variance) were used to characterize the different equations obtained and to determine the best calibration equation (Shenk and Westerhaus, 1995). To evaluate prediction effects among different amino acids, the ratio SD/SECV was introduced by dividing the value of SD by that of SECV. The prediction effect with a larger value of SD/SECV is better than that with a lower value of SD/SECV. Thus, value of SD/SECV was regarded as a criterion for evaluating the performance of calibrations. 3. Results 3.1. Variability for amino acid composition of the rapeseed meal in the calibration set The NIRS calibration set included 226 intact rapeseed samples. Seventeen different amino acids in each sample were detected by using the HCl hydrolysis-HPLC method. The mean, standard deviation and range of individual amino acids and total amino acid content of rapeseed meal were listed (Table 1). The absolute contents of the 17 amino acids in this population varied widely with a larger coefficient of variation (CV) value, from 156.50 to 585.30 g/kg. For example, lysine content and aspartic acid had ranges from 4.48 to 29.25 g/kg (mean = 21.94 g/kg; SD = 5.95 g/kg, CV = 271.20 g/kg) and 14.02 to 37.98 g/kg (mean = 26.57; SD = 4.49 g/kg, CV = 169.00 g/kg), respectively. It covered a variation of cultivars in the agricultural
114
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
Table 1 Amino acid composition of the rapeseed meal for the NIRS calibration set (n = 226). Constituent
Absolute content (g/kg) Min
Aspartic acid Threonine Serine Glutamic acid Glycine Alanine Cysteine Valine Methionine Isoleucine Leucine Tyrosine Phenylalanine Lysine Histidine Arginine Proline TAAf
a
14.02 6.36 6.33 26.98 6.68 8.11 0.16 15.99 1.54 5.25 10.92 4.39 7.78 4.48 2.31 7.02 5.43 146.96
Max
b
37.98 21.03 21.33 106.86 27.68 23.30 17.00 55.78 8.52 19.96 35.86 14.65 20.82 29.25 13.68 32.51 35.86 493.56
Relative content (g/kg TAA) c
d
RSDr
e
Min
Max
9.03 9.63 12.85 9.67 10.28 20.71 67.83 17.82 61.68 9.82 9.78 30.43 11.87 9.57 11.20 10.26 15.73 10.33
65.32 32.54 29.97 154.03 32.10 44.34 0.97 49.49 5.03 26.06 51.95 18.72 35.28 28.09 11.61 43.19 34.37 -
102.92 50.78 57.09 245.79 67.58 71.11 44.08 140.98 21.31 49.30 83.64 36.24 56.51 74.27 34.00 104.59 142.04 -
Mean
SD
Sr
26.57 15.58 15.57 75.51 18.48 17.38 5.16 26.93 5.35 14.25 25.56 9.86 15.16 21.94 9.82 23.39 20.98 347.49
4.49 3.20 3.23 17.8 4.25 2.72 3.02 10.25 1.45 2.91 5.21 2.09 2.75 5.95 2.95 5.01 4.69 67.64
0.24 0.15 0.20 0.73 0.19 0.36 0.35 0.48 0.33 0.14 0.25 0.30 0.18 0.21 0.11 0.24 0.33 3.59
Mean 77.43 44.75 44.88 215.34 52.72 50.86 14.06 80.60 15.29 41.01 73.46 28.42 43.92 61.89 27.49 67.21 60.67 -
SD 7.52 2.91 4.30 17.97 4.27 6.38 7.51 32.27 2.45 2.95 4.09 2.81 2.91 9.60 5.03 6.87 12.55 -
Sr was the partial repeatability SD of wet chemical measurement between replications (measured in g/kg) after HCL hydrolysis. a Minimum. b Maximum. c Standard deviation. d The repeatability SD. e The relative standard deviation (RSD) for repeatability measured in g/kg. f Total of amino acids.
production system and the intermediate germplasm material, therefore, the set was suitable for NIRS calibration. In order to save labor, repetition started at evaporation of dilution. Most of Sr (the repeatability standard deviation) in wet chemical measure after evaporation was well controlled, especially for 11 amino acids (aspartic acid, threonine, serine, glutamic acid, glycine, isoleucine, leucine, phenylalanine, lysine, histidine, and arginine) with relatively low Sr (0.11–0.73 g/kg) and RSDr (9.10–13.00 g/kg). But, there were relatively high Sr (0.30–0.48 g/kg) and RSDr (15.60–67.80 g/kg) for 6 amino acids (cysteine, methionine, tyrosine, alanine, proline, and valine) in wet chemical analysis. In order to assess rapeseed protein quality, the relative content of each amino acid was calculated by the absolute content for each sample. The means of the relative amino acids contents ranged from 15.29 to 215.34 g/kg and showed high differences among this population and the relative amino acid contents also showed wide variability with high SD values from 2.45 to 32.27 g/kg. 3.2. Spectral analysis for amino acids The whole spectral region (1100–2498 nm) was used for the development of calibration equations. Although NIRS procedure is inherently empirical, its underlying analytical basis can be explored by identifying the spectral regions associated with the analyzed trait. When the scatter correction method (SNV + D) and math treatment with second-derivative were employed to the whole spectrum region of 1100–2498 nm, spectral absorption regions with higher correlation coefficients are apparent (Fig. 1). The correlation coefficients among the spectral regions 1136–1270, 1318–1456, 1458–1562, 1654–1786, 1830–1964, 2034–2194, 2256–2376 nm were 0.864, −0.849, −0.897, 0.807, −0.820, 0.876, 0.764, respectively. Seven wavelength regions were found in significant variation among the intact rapeseed samples due to their large SD values. All the wavelength regions contributed to the calibration equation for the amino acids because of large absolute values of correlation coefficient in these regions. But the frontal 6 regions showed mainly contribution due to their high correlation coefficient (absolute values: 0.801–0.897). 3.3. The development of NIRS calibration models for amino acids contents Prediction equations for individual and total amino acid contents in intact rapeseeds were developed using the modified partial least squares (MPLS) regression while standard normal variate and detrending transformations (SNV + D) were applied to correct scattering. Two mathematical treatments of (1,4,4,1) and (2,4,4,1) where each number in the brackets stands for derivative, gap, smooth and second smooth, respectively, were tested on the calibration set. When the mathematical treatment of (1,4,4,1) was used, good equations were obtained for the 16 amino acids except for valine (Table 2). 12 amino acids (aspartic acid, threonine, serine, glutamic acid, glycine, isoleucine, leucine, tyrosine, phenylalanine, lysine, histidine and arginine) showed high coefficients of determination with R2 from 0.89 to 0.98 and 1 − VR from 0.88 to 0.97 and high SD/SECV ratios (from 2.92 to 5.90). There were obvious effects of first derivative pretreatment. The general effect of calibration equations was good with the means of SECV, 1 − VR and SD/SECV being 2.31 g/kg, 0.88 and
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
115
Fig. 1. Standard deviations (SD) with second-derivative math treatment and the total amino acid correlation on wavelength. Dashed curve (upper) is correlation coefficient and real one (down) is SD.
3.50, respectively. The calibration equation for valine content was inadequate for prediction even though the parameters of SECV, 1 − VR and SD/SECV improved to 5.97 g/kg, 0.53 and 1.45, respectively. The information in Table 2 further showed the effects of calibration equations developed by using mathematical treatment of (2,4,4,1). When using this mathematical treatment, the equations for some amino acids were improved, especially those for 5 amino acids (aspartic acid, threonine, serine, alanine, and valine), although those for methionine, isoleucine, tyrosine, phenylalanine, cysteine, and proline were not improved as compared to using the treatment of (1,4,4,1). The general effect for amino acid calibration equations was improved with the means of SECV, 1 − VR and SD/SECV of 2.26 g/kg, 0.88 and 3.55. Therefore, the effect of second derivative mathematical treatments was slightly superior to that of first derivative mathematical treatments. The equations for most of amino acids, except for alanine, cysteine, valine, methionine and proline, showed high determination coefficients (R2 values in calibration and 1 − VR in cross-validation), and low SEC and SECV. These 12 different amino acids could be predicted with very good precision (R2 with 0.88–0.98, 1 − VR with 0.86–0.97, and SD/SECV with 2.69–5.90). The equations for 3 different amino acids (alanine, proline, and methionine) showed better agreement between the reference values and NIRS spectra with R2 (0.80–0.84) and 1 − VR (0.78–0.82). Whereas, the calibration model for cysteine (R2 = 0.76 and 1 − VR = 0.74) and valine (R2 = 0.62 and 1 − VR = 0.58) explained less variation, and predicted the amino acid with less accuracy. The NIRS prediction equation for total amino acid also showed high coefficient of determination and SD/SECV of 0.93 and 3.87, and lower SEC and SECV of 15.49 and 17.01 g/kg, respectively (Table 2). On the basis of the requirements for balance or growth from a diet containing nitrogen, AA are traditionally classified as nutritionally essential (indispensable) or non-essential (dispensable) for humans and animals. Calibration equations were successfully established for the contents of five essential amino acids (threonine, lysine, leucine, isoleucine and phenylalanine). These five essential amino acids could be predicted with good precision (R2 from 0.91 to 0.97; 1 − VR from 0.90 to 0.96; and SD/SECV from 3.13 to 5.13) (Table 2). Besides synthesizing proteins and polypeptides, some amino acids are important regulators of the main metabolic pathways which are essential for maintenance, growth, reproduction, and immunity in animal body. For this reason(s), they are also called functional amino acids based on the functional needs (e.g., reproduction and disease prevention) they fulfill, and they include arginine, cysteine, glutamine, leucine, proline, and tryptophan (Wu, 2009). Calibration equations were successfully developed for three functional amino acids content (arginine, glutamine, and leucine). These three functional amino acids could be predicted with better accuracy as shown in Table 2 (R2 ranged from 0.96 to 0.97, 1 − VR from 0.95 to 0.96, and SD/SECV from 4.50 to 4.80). 3.4. Calibration equations of relative content of amino acids Table 3 showed calibration statistics of the relative contents of amino acids with same sample set using mathematical treatment of (2,4,4,1). On the whole, the equations for relative contents of amino acids did not show effective calibration and cross-validation when compared with those equations based on absolute contents. Eleven different amino acids equations (aspartic acid, glutamic acid, glycine, alanine, cysteine, valine, leucine, phenylalanine, lysine, histidine and arginine) could be used in the prediction because their R2 values (0.72–0.95), 1 − VR (0.72–0.94) and SD/SECV (1.80–3.95) were higher and their SECs and SECVs were lower. For the other 6 amino acids, their R2 values, 1 − VR and SD/SECV were low in calibration and cross-validation making their equations unacceptable for analysis. Although, calibration equations of relative contents for most amino acids were poorer than those for absolute content with the same set (Table 3), the equations for relative
116
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
Table 2 Calibration and cross-validation statistics in NIRS equations for absolute contents of amino acids in intact rapeseeds with the math treatments of (1,4,4,1) (A) and (2,4,4,1) (B).a Predicted from TAA (R2 )
Math treatments
Aspartic acid
0.83
Threonine
0.92
Serine
0.81
Glutamic acid
0.93
Glycine
0.94
Alanine
0.73
Cysteine
0.40
Valine
0.04
Methionine
0.68
Isoleucine
0.91
Leucine
0.95
Tyrosine
0.79
Phenylalanine
0.94
Lysine
0.85
Histidine
0.89
Arginine
0.84
Proline
0.59
A B A B A B A B A B A B A B A B A B A B A B A B A B A B A B A B A B A B
Constituent
TAAg Mean a b c d e f g
Calibration
Cross-validation
N
Mean
SD
Min
Max
SEC
R
SECVe
1 − VRf
SD/SECV
218 216 214 214 211 210 214 214 221 221 217 218 214 214 211 207 220 219 216 217 219 220 217 222 216 217 214 214 221 219 213 217 219 216 217 216
26.51 26.65 15.60 15.63 15.77 15.78 76.18 75.99 18.50 18.54 17.28 17.29 4.77 4.77 25.38 25.02 5.41 5.39 14.32 14.34 25.58 25.65 9.89 9.89 15.30 15.30 22.15 22.03 9.87 9.85 23.37 23.45 20.90 20.98 346.31 345.83
4.41 4.33 3.22 3.23 3.15 3.17 17.27 17.46 4.19 4.23 2.63 2.68 2.44 2.44 8.66 8.31 1.40 1.39 2.90 2.88 5.21 5.12 2.05 2.07 2.66 2.66 5.79 5.96 2.91 2.92 4.99 4.94 4.60 4.52 66.74 66.40
13.27 13.66 5.94 5.94 6.31 6.28 24.39 23.60 5.94 5.84 9.39 9.24 0.00 0.00 0.00 0.10 1.21 1.23 5.62 5.69 9.95 10.29 3.75 3.69 7.32 7.33 4.77 4.15 1.15 1.11 8.39 8.62 7.11 7.42 146.09 146.63
39.75 39.64 25.27 25.32 25.22 25.28 127.98 128.38 31.07 31.24 25.18 25.34 12.08 12.08 51.36 49.95 9.61 9.54 23.02 22.99 41.20 41.02 16.02 16.10 23.29 23.27 39.53 39.90 18.59 18.60 38.35 38.28 34.68 34.53 546.54 545.03
1.18 0.95 0.65 0.61 0.91 0.83 3.36 3.34 0.74 0.74 1.13 1.08 1.19 1.22 5.62 5.12 0.60 0.62 0.82 0.85 0.99 0.94 0.68 0.73 0.65 0.70 1.03 1.00 0.46 0.46 0.95 0.93 1.84 1.96 15.28 14.93
0.93 0.95 0.96 0.96 0.92 0.93 0.96 0.96 0.97 0.97 0.81 0.84 0.76 0.75 0.58 0.62 0.82 0.80 0.92 0.91 0.96 0.97 0.89 0.88 0.94 0.93 0.97 0.97 0.98 0.98 0.96 0.97 0.84 0.81 0.95 0.95
1.31 1.18 0.68 0.68 0.94 0.90 3.76 3.73 0.78 0.81 1.16 1.14 1.25 1.27 5.97 5.41 0.65 0.64 0.88 0.92 1.09 1.09 0.70 0.77 0.70 0.74 1.19 1.16 0.51 0.49 1.12 1.20 2.08 2.11 16.74 16.51
0.91 0.93 0.96 0.96 0.91 0.92 0.95 0.95 0.97 0.96 0.81 0.82 0.74 0.73 0.53 0.58 0.78 0.78 0.91 0.90 0.96 0.95 0.88 0.86 0.93 0.92 0.96 0.96 0.97 0.97 0.95 0.94 0.79 0.78 0.94 0.94
3.36 3.68 4.72 4.79 3.37 3.51 4.59 4.68 5.34 5.22 2.27 2.37 1.94 1.92 1.45 1.54 2.15 2.15 3.29 3.13 4.80 4.69 2.92 2.69 3.79 3.58 4.86 5.13 5.72 5.90 4.48 4.11 2.21 2.14 3.99 4.02
38.50 38.47
8.07 8.04
14.48 14.49
62.71 62.58
2.12 2.05
0.90 0.90
2.31 2.26
0.88 0.88
3.50 3.55
A B
b
c
2d
Scatter correction: SNV + D and regression method: MPLS. Standard deviation. The standard error of calibration. Coefficient of determination in calibration. Standard error of cross-validation. 1 minus the ratio of unexplained variance to total variance. Total of amino acids.
contents of 2 amino acids (alanine and valine), unlike most amino acids, were better than those from their absolute content because the former had higher parameters of R2 (0.93, 0.82), 1 − VR (0.89, 0.81), and the ratio SD/SECV (2.94, 2.29) comparing between Tables 2 and 3. The relative content of amino acids was a very important index for estimating protein quality in intact seed. Although no satisfactory results were acquired in the calibration equations for most amino acids (Table 3), 9 amino acids (aspartic acid, glutamic acid, glycine, alanine, valine, leucine, lysine, histidine, and arginine) could be predicted with relative accuracy. Fortunately, there were good calibration equations for absolute content of 16 amino acids in rapeseed meal without loss of precision. Thus the relative content could be calculated after its absolute contents were predicted by NIRS. 4. Discussion The amino acids are vital for the promotion of muscle growth, milk production, egg, and meat quality, while avoiding excess fat deposition (Wu, 2009). Some amino acids, such as lysine and leucine, are limiting amino acids in rapeseed meal for animals while other amino acids including arginine, cysteine, glutamine, and proline are functional amino acids. Adding one or a mixture of these amino acids could help to cure diseases at various stages of the life cycle, enhancing the efficiency
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
117
Table 3 Calibration and cross-validation statistics in NIRS equationsa for relative contents of amino acids in intact rapeseeds with math treatment (2,4,4,1). Constituent
Aspartic acid Threonine Serine Glutamic acid Glycine Alanine Cysteine Valine Methionine Isoleucine Leucine Tyrosine Phenylalanine Lysine Histidine Arginine Proline Mean a b c d e f
Calibration
Cross-validation
N
Mean
SD
Min
Max
SEC
R
SECVe
1 − VRf
SD/SECV
222 220 209 214 216 212 215 209 218 213 218 219 213 215 218 215 216
77.27 44.88 45.48 215.51 52.87 50.4 13.25 77.76 15.45 41.31 73.64 28.45 44.08 62.64 27.92 67.13 59.21 58.66
7.41 2.72 3.09 17.63 3.55 6.07 6.12 31.87 2.15 2.41 3.81 2.59 2.61 8.96 4.56 6.30 7.84 7.04
55.04 36.73 36.23 162.63 42.24 32.20 0.00 0.00 9.01 34.07 62.20 20.68 36.25 35.77 14.24 48.23 35.70 38.90
99.49 53.03 54.74 268.39 63.50 68.60 31.60 173.35 21.89 48.54 85.08 36.23 51.90 89.51 41.60 86.03 82.71 79.78
3.16 1.89 2.63 6.45 1.43 1.60 3.26 13.43 1.36 1.46 1.72 2.09 1.25 2.74 1.00 2.32 4.04 3.05
0.82 0.52 0.27 0.87 0.84 0.93 0.72 0.82 0.60 0.63 0.80 0.35 0.77 0.91 0.95 0.86 0.73 0.73
3.57 2.01 2.80 7.38 1.68 2.06 3.39 13.92 1.61 1.85 2.08 2.20 1.38 3.06 1.15 3.16 5.85 3.48
0.77 0.45 0.17 0.83 0.78 0.89 0.69 0.81 0.43 0.41 0.70 0.27 0.72 0.88 0.94 0.75 0.44 0.64
2.08 1.35 1.10 2.39 2.11 2.94 1.80 2.29 1.33 1.30 1.83 1.18 1.90 2.93 3.95 1.99 1.34 1.99
b
c
2d
Scatter correction: SNV + D and regression method: MPLS. Standard deviation. The standard error of calibration. Coefficient of determination in calibration. Standard error of cross-validation. 1 minus the ratio of unexplained variance to total variance.
of metabolic transformations. Thus, increasing the content of limiting and functional amino acids is very important for improving rapeseed meal’s nutritional value. It is necessary to understand the mechanism of gene expression on protein and amino acids content across environments. The genetic control of seed protein and amino acids content in rapeseed is complicated. Grami and Stefansson (1977) and Grami et al. (1997) found that protein trait in rapeseed is strongly affected by the environment, exhibiting a low heritability. However, the results of a study by Wang and Qiu (1990) showed that the broad- and narrow-sense heritability of protein content was as high as 0.748 and 0.572, respectively. Recently, some results found that protein and amino acid contents of rapeseed were simultaneously controlled by genetic effects of embryo, cytoplasm, and maternal plant, of which the maternal genetic effects were most important, followed by embryo and cytoplasmic genetic effects. The total narrow-sense heritability for protein content was 64.17%. Therefore, improvement of protein and amino acid contents of rapeseed would be more efficient when selection is based on maternal plants than that on single seeds. Selection for improving protein content can therefore be conducted in early generations (Ren et al., 2005; Wu et al., 2005; Variath et al., 2009). In addition, rapeseed protein and amino acid content could be changed by molecular techniques (Nesi et al., 2008). Modification of protein and amino acid contents in rapeseed is feasible because of large variation in amino acids such as lysine content. So, it is more necessary to construct a nondestructive, fast, reliable and robust method for screening intermediate lines in early generations. A small size of the standard error (SECV) does not clearly express the availability of an NIRS calibration for quality traits evaluation. SD/SECV is the ratio of the standard deviation (SD) of the amino acids content to the SECV, which was regarded as a criterion for assessing comprehensively the usefulness of NIRS predictions based on variation in the calibration set, processing of spectra data, and the control of errors from chemical measurements (Fontaine et al., 2001). If the SD/SECV ratio is high, NIRS prediction makes it easy to divide a given amount of samples into low, medium, and high contents of the amino acid. The corresponding equations could predict the contents of the amino acid with strong ability. Large variation in the calibration set is the prerequisite to achieve high SD/SECV. Small lab error in wet chemical analysis is also important and should be effectively controlled (Mentink et al., 2006). Studies by Fontaine et al. (2001, 2002) elucidated mechanism to successfully establish NIRS models. In present study, the prediction effects of most equations for relative contents of amino acids decreased due to lab errors multiplied. The SD/SECV ratios for all amino acid calibrations were shown in present project (Table 2). If this ratio exceeds a value of 3, the calibration equation is very meaningful and could be used to predict the amino acid content, values below 2 and above 1.5 have limited prediction effects, but the results may be accepted in a breeding program. In this study the (SD/SECV) ratio values of >3 were obtained for most calibration equations. Eleven different amino acids could be predicted with very good precision (R2 from 0.91 to 0.98, 1 − VR from 0.90 to 0.97 and SD/SECV with 3.13 to 5.90). The best results were derived for histidine, glycine and lysine with the SD/SECV of 5.90, 5.22, and 5.13, respectively. Their corresponding R2 were above 0.97. The equations for contents of alanine showed better agreement between reference value and spectra by NIRS with moderate determination coefficients (R2 = 0.84, 1 − VR = 0.82 and SD/SECV = 2.37). The calibration model for valine (R2 = 0.62, 1 − VR = 0.58 and SD/SECV = 1.54) might explain less variation, and predict the amino acid with less accuracy, but could be employed qualitatively in breeding. The NIRS prediction equations
118
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
for total amino acid content also showed high coefficient of determination and SD/SECV, and low SEC and SECV. Therefore, calibration equations were successfully developed for 16 different amino acids, except for model of valine content which gave a low coefficient of determination in the calibration. These calibration equations showed high accuracy in the prediction of amino acid contents of rapeseed meal. The results of NIRS calibration for amino acids in rapeseed meal were first reported with intact rapeseeds. When compared with the absolute contents, the relative contents of amino acids had poorer effects on the calibration equations. Each individual amino acid has lab error and the total AA has lab error too. When the relative content came from the individual absolute content divided by TAA, thus, it is lab error divided by error and the relative content error goes up × fold. So the relative contents all suffer from dual analytical error and lead to worse NIRS calibration effects. NIRS can predict amino acids contents in cereals much better than crude protein regressions, especially for lysine and methionine (Fontaine et al., 2001, 2002). In the present study, the result of correlation between individual amino acid content and TAA displayed the proportion of the variation explained merely by the TAA (Table 2). Most proportions between individual amino acids and TAA were smaller than R2 in calibration or 1 − VR in cross-validation for absolute contents. Additionally, all the relative contents except for serine and tyrosine show high R2 in calibration, whose the calibration effects of absolute content are associated with both individual amino acid and TAA (Table 3). Because of the bonds constitute, all NIR spectra of the amino acids were associated with the common bonds and specific bonds. The calibration of 14 individual amino acids contributed significantly, which suggested that their calibrations are promising, except for those for isoleucine, leucine and phenylalanine (R2 with 0.92–0.96) because isoleucine, leucine and phenylalanine highly correlate with TAA (R2 with 0.91–0.95). However considering the advantages of NIR technology, the calibrations on these three individual amino acid contents are still essential. The results indicated that the smaller the correlation coefficient between an individual amino acid content and TAA, the more reliable the calibration equation becomes. Calibration models are superior to traditional correlation coefficient method in prediction of content of amino acids in rapeseed meal. The NIRS analysis with intact rapeseeds offers an advantage of avoiding grinding or other sample preparations. The results of this study showed that the NIRS calibrations for amino acids could be directly used in breeding programs. In addition, one of the main advantages of this technique is that the content of amino acids are simultaneously analyzed along with the fatty acid content, and other traits, such as glucosinolate content and oil content of rapeseed. Therefore, these results demonstrated that NIRS is a powerful tool for nondestructive assessment of variation for amino acid contents in rapeseed. NIRS as a cost-effective and useful analytical tool has considerable potential in the rapeseeds improvement program for the identification and screening of large numbers of breeding materials in the early stage of breeding. 5. Conclusion Optimizing sample presentation and combination of data processing are very important for reliable and excellent equations. Combination of SNV + D/“2,4,4,1”/MPLS/was the best method in this study. Future research should aim at enhancing the accuracy of the prediction by adding new samples to the calibration set, reducing the lab error of the wet chemistry, and optimizing spectra data processing. Recalibration and further optimization of the calibration equations will probably yield an improvement in their performance, and make them more suitable on large scale in rapeseed breeding. The purpose of developing these NIRS equations was to obtain a rapid method for estimating individual and total amino acid contents based on intact rapeseeds. The results exhibited that NIRS can be used as a screening method to evaluate a large number of samples in a short period of time. The NIRS calibration equations developed in this study allow for simultaneous analysis along with other quality traits of rapeseed in a nondestructive, high-speed and cost-effective way. Acknowledgements This research was financially supported from Technology Office of Zhejiang Province (No. 2008C22084), Foundation for University Key Teacher by the Ministry of Education of China and by the 151 Program for the Talents of Zhejiang Province. We thank Mr. Yuanlong Li, the technician of 985-Institute of Agribiology and Environmental Sciences of Zhejiang University, for the assistance in using the L-8900 amino acid analyzer. References Abe, H., Kusama, T., Kawano, S., Iwamoto, M., 1995. Nondestructive determination of amino acid content in a single kernel of wheat and soybean by near infrared spectroscopy. In: Davies, A.M.C., Williams, P. (Eds.), Near Infrared Spectroscopy: The Future Waves. NIR Publications, Chichester, UK, pp. 457–461. AOAC, 1980. Official Methods of Analysis of the Association of Official Analytical Chemists, 13th edition. The Association, Washington, DC. Barnes, R.J., Dhanoa, M.S., Lister, S.J., 1989. Standard normal variant transformation and de-trending of near-infrared reflectance spectra. Appl. Spectrosc. 43, 772–777. Baye, T.M., Pearson, T.C., Settles, A.M., 2006. Development of a calibration to predict maize seed composition using single kernel by near infrared spectroscopy. J. Cereal. Sci. 43, 236–243. Ben-Gera, I., Norris, K.H., 1968. Determination of moisture content in soybeans by direct spectrophotometry. Israelic J. Agric. Res. 18, 125–132. Downey, R.K., Harvey, B.L., 1963. Methods of breeding for oil quality in rape. Can. J. Plant Sci. 43, 271–275. Fontaine, J., Hörr, J., Schirmer, B., 2001. Near-infrared reflectance spectroscopy enables the fast and accurate prediction of the essential amino acid contents in soy, rapeseed meal, sunflower meal, peas, fishmeal, meat meal products, and poultry meal. J. Agric. Food Chem. 49, 57–66.
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
119
Fontaine, J., Schirmer, B., Hörr, J., 2002. Near-infrared reflectance spectroscopy (NIRS) enables the fast and accurate prediction of essential amino acid contents. 2. Results for wheat, barley, corn, triticale, wheat bran/middlings, rice bran, and sorghum. J. Agric. Food Chem. 50, 3902–3911. Goding, L.A., Downey, R.K., Finlayson, A.J., 1972. Seed protein amino acid compositions resulting from crosses between two Brassica campestris cultivars. Can. J. Plant Sci. 52, 63–71. Grami, B., Baker, R.J., Stefansson, B.R., 1997. Genetics of protein and oil content in summer rape: heritability, number of effective factors and correlations. Can. J. Plant Sci. 57, 937–943. Grami, B., Stefansson, B.R., 1977. Gene action for protein and oil content in summer rape. Can. J. Plant Sci. 57, 625–631. Hom, H.H., Heiko, C., Möllers, C., 2007. Non-destructive analysis of rapeseed quality by NIRS of small seed samples and single seeds. Euphytica 153, 27–34. Huisman, J., Tolman, G.H., 1990. Antinutritional factors in the plant proteins of diets for non-ruminants. In: Haresign, M., Cole, D.J.A. (Eds.), Recent Advances in Animal Nutrition. Butterworth–Heinemann, Oxford, pp. 3–31. Josefesson, E., Muhlenberg, C., 1968. Glucosinolates in seed of Swedish crops. Acta Agric. Scand. 18, 97–100. Mentink, R.L., Hoffman, P.C., Bauman, L.M., 2006. Utility of near-infrared reflectance spectroscopy to predict nutrient composition and in vitro digestibility of total mixed rations. J. Dairy Sci. 89, 2320–2326. Nesi, N., Delourme, R., Brégeon, M., Falentin, C., Renard, M., 2008. Genetic and molecular approaches to improve nutritional value of Brassica napus L. seed. CR Soc. Biol. 331, 763–771. Pujol, S., Pérez-Vendrell, A.M., Torrallardona, D., 2007. Evaluation of prediction of barley digestible nutrient content with near-infrared reflectance spectroscopy (NIRS). Livest. Sci. 109, 189–192. Ren, Y.L., Wu, J.G., Shi, C.H., 2005. Analysis of embryo, cytoplasmic and maternal effects on three amino acid traits in rapeseed. J. Zhejing Univ. (Agricultural and Life Science) 31, 41–46. Röbbelen, G., 1981. Potential and restrictions of breeding for amino acid improvement in rapeseed. In: Bunting, E.S. (Ed.), Production and Utilization of Amino acid in Oilseed Crops. Martinus Nijhoff, The Netherlands, pp. 3–11. Roussel, V., Branlard, G., Vézine, J.C., Bertrand, D., Balfourier, F., 2005. NIRS analysis reveals temporal trends in the chemical composition of French bread wheat accessions cultivated between 1800 and 2000. J. Cereal. Sci. 42, 193–203. Shenk, J.S., Westerhaus, M.O., 1991a. Population definition, sample selection, and calibration procedures for near infrared spectroscopy. Crop Sci. 31, 469–474. Shenk, J.S., Westerhaus, M.O., 1991b. Population structuring of near infrared spectra and modified partial least squares regression. Crop Sci. 31, 1548–1555. Shenk, J.S., Westerhaus, M.O., 1993. Analysis of Agriculture and Food Products by Near Infrared Reflectance Spectroscopy. Infrasoft International, Port Matilda, USA. Shenk, J.S., Westerhaus, M.O., 1995. Routine Operation, Calibration Development and Network System Management Manual for Near Infrared Instruments Version 3. 1. Infrasoft International, Port Matilda, USA. Simbaya, J., Alominski, S.B., Rakow, G., Downey, R.K., 1995. Quality characteristics of yellow-seeded Brassica seed meals: protein, carbohydrates, and dietary fiber components. J. Agric. Food Chem. 43, 2062–2066. Theander, O., Aman, P., Westlund, E., Andersson, R., Pettersson, D., 1995. Total dietary fiber determined as neutral sugar residues, uronic acid residues, and klason lignin (the Uppsala method): collaborative study. J. AOAC Int. 78, 1030–1044. Variath, M.T., Wu, J.G., Li, Y.X., Chen, G.L., Shi, C.H., 2009. Genetic analysis for oil and protein contents of rapeseed (Brassica napus L.) at different developmental times. Euphytica 166, 145–153. Velasco, L., Becker, H.C., 1998a. Analysis of total glucosinolate content and individual glucosinolates in Brassica spp. by near-infrared reflectance spectroscopy. Plant Breed. 117, 97–102. Velasco, L., Becker, H.C., 1998b. Estimating the fatty acid composition of the oil in intact-seed rapeseed (Brassica napus L.) by near-infrared reflectance spectroscopy. Euphytica 101, 221–230. Velasco, L., Fernández, J.M., De Haro, A., 1997. Determination of the fatty acid composition of the oil in intact-seed mustard by near-infrared reflectance spectroscopy. J. Am. Oil Chem. Soc. 74, 1595–1602. Velasco, L., Möllers, C., 2002. Nondestructive assessment of protein content in single seeds of rapeseed (Brassica napus L.) by near-infrared reflectance spectroscopy. Euphytica 123, 89–93. Velasco, L., Möllers, C., Becker, H.C., 1999. Estimation of seed weight, oil content and fatty acid composition in intact single seeds of rapeseed (Brassica napus L.) by near infrared reflectance spectroscopy. Euphytica 106, 79–85. Wang, F., Qiu, J., 1990. Studies on the inheritance of seed protein content and its correlation with other characters in Brassica napus L. Sci. Agric. Sin. 23, 42–47. Wu, G.Y., 2009. Amino acids: metabolism, functions, and nutrition. Amino Acids 37, 1–17. Wu, J.G., Shi, C.H., 2004. Prediction of grain weight, brown rice weight and amylose content in single rice grains using near-infrared reflectance spectroscopy. Field Crops Res. 87, 13–21. Wu, J.G., Shi, C.H., Zhang, H.Z., 2005. Genetic analysis of embryo, cytoplasmic, and maternal effects and their environment interactions for protein content in Brassica napus L. Aust. J. Agric. Res. 56, 69–73. Wu, J.G., Shi, C.H., Zhang, H.Z., Fan, L.J., 2002a. Study on analysis of oil content in intact rapeseed with few sample size by near infrared reflectance spectroscopy. Acta Agric. Sin. 28, 421–425. Wu, J.G., Shi, C.H., Zhang, X.M., 2002b. Estimating the amino acid composition in the milled rice powder by near-infrared reflectance spectroscopy. Field Crops Res. 75, 1–7.
Contents lists available at ScienceDirect
Animal Feed Science and Technology journal homepage: www.elsevier.com/locate/anifeedsci
Nondestructive assessment of amino acid composition in rapeseed meal based on intact seeds by near-infrared reflectance spectroscopy G.L. Chen a,b , B. Zhang a , J.G. Wu a,∗ , C.H. Shi a,∗ a b
Department of Agronomy, Zhejiang University, Hangzhou 310029, China School of Agriculture and Food Science, Zhejiang Forestry University, Linan 311300, China
a r t i c l e
i n f o
Article history: Received 20 January 2010 Received in revised form 25 January 2011 Accepted 4 February 2011
Keywords: Near infrared reflectance spectroscopy (NIRS) Amino acids content Rapeseed meal Intact rapeseed
a b s t r a c t The ability of near-infrared spectroscopy (NIRS) was tested for estimating individual and total amino acid contents in rapeseed meal. Twelve different amino acids (aspartic acid, threonine, serine, glutamic acid, glycine, isoleucine, leucine, tyrosine, phenylalanine, lysine, histidine and arginine) in rapeseed meal could be predicted. The R2 ranged from 0.89 to 0.98, 1 − VR (1 minus the ratio of unexplained variance to total variance) ranged from 0.86 to 0.97 and the ratio of sample standard deviation (SD) to the standard error of cross-validation (SECV) ranged from 2.69 to 5.90. The equations for alanine showed better agreement between reference value and spectra (R2 = 0.84, 1 − VR = 0.82 and SD/SECV = 2.15). The calibration model for proline (R2 = 0.81 and 1 − VR = 0.78), methionine (R2 = 0.80 and 1 − VR = 0.78), cysteine (R2 = 0.76 and 1 − VR = 0.74), and valine (R2 = 0.62 and 1 − VR = 0.58) however explained less variation. The NIRS prediction equation for total amino acid also showed high coefficient of determination (R2 = 0.93) and SD/SECV (3.87), and low SECV (17.01 g/kg). Equations of 9 amino acids (aspartic acid, glutamic acid, glycine, alanine, valine, leucine, lysine, histidine and arginine) were developed for relative contents of total amino acid and deemed useful for prediction with R2 values from 0.80 to 0.95, 1 − VR from 0.70 to 0.95 and SD/SECV from 1.83 to 3.95 and reasonably low SECV values. These results demonstrated that NIRS is a reliable tool for nondestructive assessment of variation in amino acid contents, increasing the efficiency of breeding and accelerating the selection process in rapeseed. © 2011 Elsevier B.V. All rights reserved.
1. Introduction Brassica oilseed crops are mainly grown for the production of vegetable oils which are used for human consumption or industrial applications and their byproduct are high protein feed meal which includes about 400 g/kg of crude protein. Rapeseed protein which has a rational amino acid composition is another important source of nutrition (Goding et al., 1972; Huisman and Tolman, 1990). The contents of two amino acids (methionine and cysteine) are higher than those in soybean and peanut. Soluble protein, lysine and other essential amino acid contents are also higher than those in sunflower and
Abbreviations: 1 − VR, 1 minus the ratio of unexplained variance to total variance; CV, the coefficient of variation; MPLS, modified partial least squares; NIRS, near infrared reflectance spectroscopy; R2 , coefficient of determination in calibration; RSDr , relative standard deviation for repeatability; SD, standard deviation; SD/SECV, ratio of the standard deviation (SD) of the amino acid contents in the calibration samples to the standard error of cross-validation (SECV); SEC, the standard error of calibration; SECV, standard error of cross-validation; SNV + D, standard normal variate + detrending; Sr , repeatability of standard deviation; TAA, total of amino acids. ∗ Corresponding authors. Tel.: +86 571 86971691; fax: +86 571 86971117. E-mail addresses: [email protected] (J.G. Wu), [email protected] (C.H. Shi). 0377-8401/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.anifeedsci.2011.02.004
112
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
sesame. The nutritional value of rapeseed meal is not less than that of soybean (Josefesson and Muhlenberg, 1968; Simbaya et al., 1995). Wu et al. (2005) found that protein content of rapeseed was simultaneously controlled by genetic effects of embryo, cytoplasm, and maternal plant, of which the maternal genetic effect was most important. The total narrow-sense heritability is high. Röbbelen (1981) pointed out that the development of high-protein cultivars in rapeseed is an important objective for breeding because the rapeseed meal can be transformed into good feed for livestock. High protein content and rational amino acids composition in rapeseed are major concern to the breeder. Improvement for seed quality traits depends on the availability of fast and accurate methods for their measurement in breeding practice (Wu and Shi, 2004). Near infrared reflectance spectroscopy (NIRS) technique started in 1960s (Ben-Gera and Norris, 1968), has developed rapidly since late 1980s, and become a powerful green analytical tool for the routine analysis of quality traits in rapeseed. Using intact seeds sample, NIRS calibration equations have been confirmed to be useful to routinely determine simultaneously several quality traits, such as oil, protein and glucosinolate contents as well as fatty acid composition (Velasco et al., 1997; Velasco and Becker, 1998a,b; Velasco et al., 1999; Wu et al., 2002a). Velasco and Möllers (2002) reported that the NIRS technique was reliable enough for the estimation of protein content in a single rapeseed. Hom et al. (2007) compared the accuracy of NIRS analysis of protein and oil content in intact seeds sample size of 3–4 g with the same content in a single seed (5 mg), and concluded that standard NIRS calibration equations could be used for screening a single rapeseed. The method to determine the fatty acid composition of the seed oil at a half-seed level (Downey and Harvey, 1963) has been substituted by the above method for the improvement of seed oil quality. Near-infrared spectroscopy has been used on intact grains of wheat (Abe et al., 1995; Roussel et al., 2005), milled rice flour (Wu et al., 2002b) and a single corn kernel (Baye et al., 2006) to predict nutrient quality traits such as protein and amino acid content. In the feed industry, quick and accurate analysis methods on contents of essential amino acids in the most important protein-rich feed draws a lot of attention from feed researchers. NIRS calibrations were developed for the accurate and fast determination of the total contents of important essential amino acids and protein with the finely ground material in the most cereals, bran and meals for animal feed production (Fontaine et al., 2001, 2002). Their equations facilitate the routine work, improve the accuracy of feed formulation and quality, and decrease production cost (Pujol et al., 2007). NIRS technique is widely applied in agricultural product analyses and breeding programs (Wu et al., 2002a,b). Protein content is determined in the laboratory by a classical procedure of the Kjeldahl method, and amino acid content by the HCl hydrolysis-HPLC method. These conventional chemical analytical methods are time-consuming and expensive. They are also considered destructive because they require grinding and other pretreatments of samples which make them unsuitable for the analysis of a large number of samples in the early generations of breeding. A rapid and accurate method to identify and screen breeding materials without sample preparation is crucial for the efficiency in breeding and an acceleration of the selection process. Considering the demands of breeding a rapeseed cultivar with high quality protein and its genetic analysis, it is necessary to develop a fast and efficient method such as near-infrared reflectance spectroscopy technology to determine amino acids composition in rapeseed meal. To date, no attempt has been made to determine amino acids content in rapeseed meal based on intact seeds by NIRS. It is for this reason that we have focused on the technology of NIRS. This work was to study the potentials of NIRS to estimate amino acid contents of rapeseed meal and to explore its applicability in identifying variability for these traits. 2. Materials and methods 2.1. Materials An original population of 621 rapeseed samples was mainly derived from genetic experiments and the breeding programs. All seed samples including F1 and F2 hybrid generations and their parents were collected in the years of 2007 and 2008. Being cultivars and breeding line across environments the seed samples had a larger variation in seed qualities making them suitable for NIRS calibration. All the raw samples were scanned to collect the NIRS spectra. 2.2. Collection of spectra WinISI II (Version1.04) software was applied to collect spectra and develop the calibration equations in this study. Whole intact rapeseed samples were scanned on a NIRSystems model 5000 monochromator (NIRSystems Inc., Silver Spring, MD, USA). Approximately 3 g of intact rapeseed samples were placed in a small ring cup of 36 mm inner diameter, and reflectance spectra (log 1/R) from 1100 to 2498 nm were recorded at 2 nm wavelength increment. Each sample was subsequently scanned 32 times in small ring cells (Shenk and Westerhaus, 1993). 2.3. Selection of calibration set A powerful approach to choose samples based on their spectra has been developed with the CENTER and SELECT algorithm by Shenk and Westerhaus (1991a). After the original population with 621 samples was scanned, the algorithm CENTER was used for the calculation of principal components and GH for the description of spectral boundaries and detection of outliers (Shenk and Westerhaus, 1991a). The number of samples reduced to 611 after outliers were excluded. Following centering of
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
113
the data, the SELECT algorithm was used for grouping the spectra into clusters with NH of 0.8. One representative spectrum from each cluster was selected for the arrangement of the calibration set of samples. The number of collection decreased to 238. After 12 samples were eliminated for possessing insufficient amount for determination, 226 samples were analyzed for amino acid composition by the reference method. Therefore, based on the spectra variation, 226 samples were selected for the development of a calibration equation for amino acid content in this study. 2.4. Determination of amino acids for rapeseed meal To develop rapeseed meals for each rapeseed sample, whole intact rapeseed samples were ground through a 1 mm screen and oil was extracted by Soxhlet method (AOAC, 1980) with crude fat analyzer model SZF-06 (Shanhai Xinjia Electronic Co. Ltd, China).About 3 g milled rapeseeds sample wrapped were submerged in 50 ml ether solvent for 3 h at 75 ◦ C. After oil removal, the remains were dried to constant weighed for 1 h at 105 ◦ C to be suitable for estimating amino acids content. About 30 mg of each sample meal was weighed with a readability of 0.01 mg and 10 ml of 6 M HCl was added. The air pressure was reduced to 7 Pa with vacuum pump. The test tube was enclosed by alcohol lamp with high temperature and placed into constant temperature box for hydrolysis 24 h at 105 ◦ C. Sample liquid was cooled, mixed, and diluted to 100 ml. Ten milliliter dilution was transferred to distillation flask which was held to a rotary evaporator (model RE-52AA, Shanghai, China) for evaporation with two replications. The acid was evaporated in water bath at 65 ◦ C and the residue was dissolved in 10 ml 0.02 M HCL buffer. The sample liquid was mixed by ultrasonic instrument and filtered with 0.45 m film. About 1 ml of each sample was injected into an auto-sampler bottle for analysis. The contents of 17 amino acids were measured for each sample using amino acid auto-analyzer model L-8900 (Hitachi High-Technologies Corporation, Japan). Above method was similar to that for determination of amino acids in milled rice (Wu et al., 2002a,b). The amount of each amino acid in the samples was calculated with reference to the standard solution (Type H, Wako pure chemical Industrial Corporation, Japan) and expressed as g/kg dried meal weight (DW). It was called absolute content for each amino acid. Total amino acid (TAA) was the sum of 17 amino acids in each sample meal. The relative content of each amino acid was expressed as g/kg TAA in each sample. All values were given on basis of a dry meal. The standard solution of each amino acid was 0.1 M. The Sr , the repeatability of standard deviation, defined as the mean of the absolute difference d2 /2N, where d is difference between between laboratory duplicates, was calculated using the following formula: Sr = two duplicates on an ith sample, and N is the number of samples. The relative standard deviation (RSDr = (Sr /mean) × 1000) for repeatability with unit of g/kg (Theander et al., 1995). 2.5. Data processing Calibration models for amino acids were developed using the platform of WinISI II (Version1.04) software. The spectral data with range from1100 to 2498 nm and an interval of 2 nm were subjected to two different mathematical treatments during the development of calibration models. For example in the notation (1,4,4,1), the first number indicates the order of the derivative of log(1/R), the second is the gap in data points over which the derivative is calculated, the third and fourth numbers refer to the number of data points used in the first and second smoothing, respectively. In addition, standard normal variate and detrending (SNV + D) were applied to correct scattering (Barnes et al., 1989). The SNV + D were designed to remove additive baseline and multiplicative signal effects resulting in a spectrum with zero mean and a variance equal to 1. The application of SNV + D to raw spectral data reduces the differences in spectra related to physical characteristics such as particle size. A modified partial least squares algorithm was used (Shenk and Westerhaus, 1991b). Calibrations were tested using cross-validation. Two rounds of outlier elimination were allowed, with samples with an ‘H’ value larger than 4 (spectral outliers) and a (Student) ‘t’ value larger than 2.5 (sample which did not fit the calibration model) being eliminated. The standard error of calibration (SEC), coefficient of determination (R2 ), standard error of cross-validation (SECV) and 1 − VR (1 minus the ratio of unexplained variance to total variance) were used to characterize the different equations obtained and to determine the best calibration equation (Shenk and Westerhaus, 1995). To evaluate prediction effects among different amino acids, the ratio SD/SECV was introduced by dividing the value of SD by that of SECV. The prediction effect with a larger value of SD/SECV is better than that with a lower value of SD/SECV. Thus, value of SD/SECV was regarded as a criterion for evaluating the performance of calibrations. 3. Results 3.1. Variability for amino acid composition of the rapeseed meal in the calibration set The NIRS calibration set included 226 intact rapeseed samples. Seventeen different amino acids in each sample were detected by using the HCl hydrolysis-HPLC method. The mean, standard deviation and range of individual amino acids and total amino acid content of rapeseed meal were listed (Table 1). The absolute contents of the 17 amino acids in this population varied widely with a larger coefficient of variation (CV) value, from 156.50 to 585.30 g/kg. For example, lysine content and aspartic acid had ranges from 4.48 to 29.25 g/kg (mean = 21.94 g/kg; SD = 5.95 g/kg, CV = 271.20 g/kg) and 14.02 to 37.98 g/kg (mean = 26.57; SD = 4.49 g/kg, CV = 169.00 g/kg), respectively. It covered a variation of cultivars in the agricultural
114
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
Table 1 Amino acid composition of the rapeseed meal for the NIRS calibration set (n = 226). Constituent
Absolute content (g/kg) Min
Aspartic acid Threonine Serine Glutamic acid Glycine Alanine Cysteine Valine Methionine Isoleucine Leucine Tyrosine Phenylalanine Lysine Histidine Arginine Proline TAAf
a
14.02 6.36 6.33 26.98 6.68 8.11 0.16 15.99 1.54 5.25 10.92 4.39 7.78 4.48 2.31 7.02 5.43 146.96
Max
b
37.98 21.03 21.33 106.86 27.68 23.30 17.00 55.78 8.52 19.96 35.86 14.65 20.82 29.25 13.68 32.51 35.86 493.56
Relative content (g/kg TAA) c
d
RSDr
e
Min
Max
9.03 9.63 12.85 9.67 10.28 20.71 67.83 17.82 61.68 9.82 9.78 30.43 11.87 9.57 11.20 10.26 15.73 10.33
65.32 32.54 29.97 154.03 32.10 44.34 0.97 49.49 5.03 26.06 51.95 18.72 35.28 28.09 11.61 43.19 34.37 -
102.92 50.78 57.09 245.79 67.58 71.11 44.08 140.98 21.31 49.30 83.64 36.24 56.51 74.27 34.00 104.59 142.04 -
Mean
SD
Sr
26.57 15.58 15.57 75.51 18.48 17.38 5.16 26.93 5.35 14.25 25.56 9.86 15.16 21.94 9.82 23.39 20.98 347.49
4.49 3.20 3.23 17.8 4.25 2.72 3.02 10.25 1.45 2.91 5.21 2.09 2.75 5.95 2.95 5.01 4.69 67.64
0.24 0.15 0.20 0.73 0.19 0.36 0.35 0.48 0.33 0.14 0.25 0.30 0.18 0.21 0.11 0.24 0.33 3.59
Mean 77.43 44.75 44.88 215.34 52.72 50.86 14.06 80.60 15.29 41.01 73.46 28.42 43.92 61.89 27.49 67.21 60.67 -
SD 7.52 2.91 4.30 17.97 4.27 6.38 7.51 32.27 2.45 2.95 4.09 2.81 2.91 9.60 5.03 6.87 12.55 -
Sr was the partial repeatability SD of wet chemical measurement between replications (measured in g/kg) after HCL hydrolysis. a Minimum. b Maximum. c Standard deviation. d The repeatability SD. e The relative standard deviation (RSD) for repeatability measured in g/kg. f Total of amino acids.
production system and the intermediate germplasm material, therefore, the set was suitable for NIRS calibration. In order to save labor, repetition started at evaporation of dilution. Most of Sr (the repeatability standard deviation) in wet chemical measure after evaporation was well controlled, especially for 11 amino acids (aspartic acid, threonine, serine, glutamic acid, glycine, isoleucine, leucine, phenylalanine, lysine, histidine, and arginine) with relatively low Sr (0.11–0.73 g/kg) and RSDr (9.10–13.00 g/kg). But, there were relatively high Sr (0.30–0.48 g/kg) and RSDr (15.60–67.80 g/kg) for 6 amino acids (cysteine, methionine, tyrosine, alanine, proline, and valine) in wet chemical analysis. In order to assess rapeseed protein quality, the relative content of each amino acid was calculated by the absolute content for each sample. The means of the relative amino acids contents ranged from 15.29 to 215.34 g/kg and showed high differences among this population and the relative amino acid contents also showed wide variability with high SD values from 2.45 to 32.27 g/kg. 3.2. Spectral analysis for amino acids The whole spectral region (1100–2498 nm) was used for the development of calibration equations. Although NIRS procedure is inherently empirical, its underlying analytical basis can be explored by identifying the spectral regions associated with the analyzed trait. When the scatter correction method (SNV + D) and math treatment with second-derivative were employed to the whole spectrum region of 1100–2498 nm, spectral absorption regions with higher correlation coefficients are apparent (Fig. 1). The correlation coefficients among the spectral regions 1136–1270, 1318–1456, 1458–1562, 1654–1786, 1830–1964, 2034–2194, 2256–2376 nm were 0.864, −0.849, −0.897, 0.807, −0.820, 0.876, 0.764, respectively. Seven wavelength regions were found in significant variation among the intact rapeseed samples due to their large SD values. All the wavelength regions contributed to the calibration equation for the amino acids because of large absolute values of correlation coefficient in these regions. But the frontal 6 regions showed mainly contribution due to their high correlation coefficient (absolute values: 0.801–0.897). 3.3. The development of NIRS calibration models for amino acids contents Prediction equations for individual and total amino acid contents in intact rapeseeds were developed using the modified partial least squares (MPLS) regression while standard normal variate and detrending transformations (SNV + D) were applied to correct scattering. Two mathematical treatments of (1,4,4,1) and (2,4,4,1) where each number in the brackets stands for derivative, gap, smooth and second smooth, respectively, were tested on the calibration set. When the mathematical treatment of (1,4,4,1) was used, good equations were obtained for the 16 amino acids except for valine (Table 2). 12 amino acids (aspartic acid, threonine, serine, glutamic acid, glycine, isoleucine, leucine, tyrosine, phenylalanine, lysine, histidine and arginine) showed high coefficients of determination with R2 from 0.89 to 0.98 and 1 − VR from 0.88 to 0.97 and high SD/SECV ratios (from 2.92 to 5.90). There were obvious effects of first derivative pretreatment. The general effect of calibration equations was good with the means of SECV, 1 − VR and SD/SECV being 2.31 g/kg, 0.88 and
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
115
Fig. 1. Standard deviations (SD) with second-derivative math treatment and the total amino acid correlation on wavelength. Dashed curve (upper) is correlation coefficient and real one (down) is SD.
3.50, respectively. The calibration equation for valine content was inadequate for prediction even though the parameters of SECV, 1 − VR and SD/SECV improved to 5.97 g/kg, 0.53 and 1.45, respectively. The information in Table 2 further showed the effects of calibration equations developed by using mathematical treatment of (2,4,4,1). When using this mathematical treatment, the equations for some amino acids were improved, especially those for 5 amino acids (aspartic acid, threonine, serine, alanine, and valine), although those for methionine, isoleucine, tyrosine, phenylalanine, cysteine, and proline were not improved as compared to using the treatment of (1,4,4,1). The general effect for amino acid calibration equations was improved with the means of SECV, 1 − VR and SD/SECV of 2.26 g/kg, 0.88 and 3.55. Therefore, the effect of second derivative mathematical treatments was slightly superior to that of first derivative mathematical treatments. The equations for most of amino acids, except for alanine, cysteine, valine, methionine and proline, showed high determination coefficients (R2 values in calibration and 1 − VR in cross-validation), and low SEC and SECV. These 12 different amino acids could be predicted with very good precision (R2 with 0.88–0.98, 1 − VR with 0.86–0.97, and SD/SECV with 2.69–5.90). The equations for 3 different amino acids (alanine, proline, and methionine) showed better agreement between the reference values and NIRS spectra with R2 (0.80–0.84) and 1 − VR (0.78–0.82). Whereas, the calibration model for cysteine (R2 = 0.76 and 1 − VR = 0.74) and valine (R2 = 0.62 and 1 − VR = 0.58) explained less variation, and predicted the amino acid with less accuracy. The NIRS prediction equation for total amino acid also showed high coefficient of determination and SD/SECV of 0.93 and 3.87, and lower SEC and SECV of 15.49 and 17.01 g/kg, respectively (Table 2). On the basis of the requirements for balance or growth from a diet containing nitrogen, AA are traditionally classified as nutritionally essential (indispensable) or non-essential (dispensable) for humans and animals. Calibration equations were successfully established for the contents of five essential amino acids (threonine, lysine, leucine, isoleucine and phenylalanine). These five essential amino acids could be predicted with good precision (R2 from 0.91 to 0.97; 1 − VR from 0.90 to 0.96; and SD/SECV from 3.13 to 5.13) (Table 2). Besides synthesizing proteins and polypeptides, some amino acids are important regulators of the main metabolic pathways which are essential for maintenance, growth, reproduction, and immunity in animal body. For this reason(s), they are also called functional amino acids based on the functional needs (e.g., reproduction and disease prevention) they fulfill, and they include arginine, cysteine, glutamine, leucine, proline, and tryptophan (Wu, 2009). Calibration equations were successfully developed for three functional amino acids content (arginine, glutamine, and leucine). These three functional amino acids could be predicted with better accuracy as shown in Table 2 (R2 ranged from 0.96 to 0.97, 1 − VR from 0.95 to 0.96, and SD/SECV from 4.50 to 4.80). 3.4. Calibration equations of relative content of amino acids Table 3 showed calibration statistics of the relative contents of amino acids with same sample set using mathematical treatment of (2,4,4,1). On the whole, the equations for relative contents of amino acids did not show effective calibration and cross-validation when compared with those equations based on absolute contents. Eleven different amino acids equations (aspartic acid, glutamic acid, glycine, alanine, cysteine, valine, leucine, phenylalanine, lysine, histidine and arginine) could be used in the prediction because their R2 values (0.72–0.95), 1 − VR (0.72–0.94) and SD/SECV (1.80–3.95) were higher and their SECs and SECVs were lower. For the other 6 amino acids, their R2 values, 1 − VR and SD/SECV were low in calibration and cross-validation making their equations unacceptable for analysis. Although, calibration equations of relative contents for most amino acids were poorer than those for absolute content with the same set (Table 3), the equations for relative
116
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
Table 2 Calibration and cross-validation statistics in NIRS equations for absolute contents of amino acids in intact rapeseeds with the math treatments of (1,4,4,1) (A) and (2,4,4,1) (B).a Predicted from TAA (R2 )
Math treatments
Aspartic acid
0.83
Threonine
0.92
Serine
0.81
Glutamic acid
0.93
Glycine
0.94
Alanine
0.73
Cysteine
0.40
Valine
0.04
Methionine
0.68
Isoleucine
0.91
Leucine
0.95
Tyrosine
0.79
Phenylalanine
0.94
Lysine
0.85
Histidine
0.89
Arginine
0.84
Proline
0.59
A B A B A B A B A B A B A B A B A B A B A B A B A B A B A B A B A B A B
Constituent
TAAg Mean a b c d e f g
Calibration
Cross-validation
N
Mean
SD
Min
Max
SEC
R
SECVe
1 − VRf
SD/SECV
218 216 214 214 211 210 214 214 221 221 217 218 214 214 211 207 220 219 216 217 219 220 217 222 216 217 214 214 221 219 213 217 219 216 217 216
26.51 26.65 15.60 15.63 15.77 15.78 76.18 75.99 18.50 18.54 17.28 17.29 4.77 4.77 25.38 25.02 5.41 5.39 14.32 14.34 25.58 25.65 9.89 9.89 15.30 15.30 22.15 22.03 9.87 9.85 23.37 23.45 20.90 20.98 346.31 345.83
4.41 4.33 3.22 3.23 3.15 3.17 17.27 17.46 4.19 4.23 2.63 2.68 2.44 2.44 8.66 8.31 1.40 1.39 2.90 2.88 5.21 5.12 2.05 2.07 2.66 2.66 5.79 5.96 2.91 2.92 4.99 4.94 4.60 4.52 66.74 66.40
13.27 13.66 5.94 5.94 6.31 6.28 24.39 23.60 5.94 5.84 9.39 9.24 0.00 0.00 0.00 0.10 1.21 1.23 5.62 5.69 9.95 10.29 3.75 3.69 7.32 7.33 4.77 4.15 1.15 1.11 8.39 8.62 7.11 7.42 146.09 146.63
39.75 39.64 25.27 25.32 25.22 25.28 127.98 128.38 31.07 31.24 25.18 25.34 12.08 12.08 51.36 49.95 9.61 9.54 23.02 22.99 41.20 41.02 16.02 16.10 23.29 23.27 39.53 39.90 18.59 18.60 38.35 38.28 34.68 34.53 546.54 545.03
1.18 0.95 0.65 0.61 0.91 0.83 3.36 3.34 0.74 0.74 1.13 1.08 1.19 1.22 5.62 5.12 0.60 0.62 0.82 0.85 0.99 0.94 0.68 0.73 0.65 0.70 1.03 1.00 0.46 0.46 0.95 0.93 1.84 1.96 15.28 14.93
0.93 0.95 0.96 0.96 0.92 0.93 0.96 0.96 0.97 0.97 0.81 0.84 0.76 0.75 0.58 0.62 0.82 0.80 0.92 0.91 0.96 0.97 0.89 0.88 0.94 0.93 0.97 0.97 0.98 0.98 0.96 0.97 0.84 0.81 0.95 0.95
1.31 1.18 0.68 0.68 0.94 0.90 3.76 3.73 0.78 0.81 1.16 1.14 1.25 1.27 5.97 5.41 0.65 0.64 0.88 0.92 1.09 1.09 0.70 0.77 0.70 0.74 1.19 1.16 0.51 0.49 1.12 1.20 2.08 2.11 16.74 16.51
0.91 0.93 0.96 0.96 0.91 0.92 0.95 0.95 0.97 0.96 0.81 0.82 0.74 0.73 0.53 0.58 0.78 0.78 0.91 0.90 0.96 0.95 0.88 0.86 0.93 0.92 0.96 0.96 0.97 0.97 0.95 0.94 0.79 0.78 0.94 0.94
3.36 3.68 4.72 4.79 3.37 3.51 4.59 4.68 5.34 5.22 2.27 2.37 1.94 1.92 1.45 1.54 2.15 2.15 3.29 3.13 4.80 4.69 2.92 2.69 3.79 3.58 4.86 5.13 5.72 5.90 4.48 4.11 2.21 2.14 3.99 4.02
38.50 38.47
8.07 8.04
14.48 14.49
62.71 62.58
2.12 2.05
0.90 0.90
2.31 2.26
0.88 0.88
3.50 3.55
A B
b
c
2d
Scatter correction: SNV + D and regression method: MPLS. Standard deviation. The standard error of calibration. Coefficient of determination in calibration. Standard error of cross-validation. 1 minus the ratio of unexplained variance to total variance. Total of amino acids.
contents of 2 amino acids (alanine and valine), unlike most amino acids, were better than those from their absolute content because the former had higher parameters of R2 (0.93, 0.82), 1 − VR (0.89, 0.81), and the ratio SD/SECV (2.94, 2.29) comparing between Tables 2 and 3. The relative content of amino acids was a very important index for estimating protein quality in intact seed. Although no satisfactory results were acquired in the calibration equations for most amino acids (Table 3), 9 amino acids (aspartic acid, glutamic acid, glycine, alanine, valine, leucine, lysine, histidine, and arginine) could be predicted with relative accuracy. Fortunately, there were good calibration equations for absolute content of 16 amino acids in rapeseed meal without loss of precision. Thus the relative content could be calculated after its absolute contents were predicted by NIRS. 4. Discussion The amino acids are vital for the promotion of muscle growth, milk production, egg, and meat quality, while avoiding excess fat deposition (Wu, 2009). Some amino acids, such as lysine and leucine, are limiting amino acids in rapeseed meal for animals while other amino acids including arginine, cysteine, glutamine, and proline are functional amino acids. Adding one or a mixture of these amino acids could help to cure diseases at various stages of the life cycle, enhancing the efficiency
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
117
Table 3 Calibration and cross-validation statistics in NIRS equationsa for relative contents of amino acids in intact rapeseeds with math treatment (2,4,4,1). Constituent
Aspartic acid Threonine Serine Glutamic acid Glycine Alanine Cysteine Valine Methionine Isoleucine Leucine Tyrosine Phenylalanine Lysine Histidine Arginine Proline Mean a b c d e f
Calibration
Cross-validation
N
Mean
SD
Min
Max
SEC
R
SECVe
1 − VRf
SD/SECV
222 220 209 214 216 212 215 209 218 213 218 219 213 215 218 215 216
77.27 44.88 45.48 215.51 52.87 50.4 13.25 77.76 15.45 41.31 73.64 28.45 44.08 62.64 27.92 67.13 59.21 58.66
7.41 2.72 3.09 17.63 3.55 6.07 6.12 31.87 2.15 2.41 3.81 2.59 2.61 8.96 4.56 6.30 7.84 7.04
55.04 36.73 36.23 162.63 42.24 32.20 0.00 0.00 9.01 34.07 62.20 20.68 36.25 35.77 14.24 48.23 35.70 38.90
99.49 53.03 54.74 268.39 63.50 68.60 31.60 173.35 21.89 48.54 85.08 36.23 51.90 89.51 41.60 86.03 82.71 79.78
3.16 1.89 2.63 6.45 1.43 1.60 3.26 13.43 1.36 1.46 1.72 2.09 1.25 2.74 1.00 2.32 4.04 3.05
0.82 0.52 0.27 0.87 0.84 0.93 0.72 0.82 0.60 0.63 0.80 0.35 0.77 0.91 0.95 0.86 0.73 0.73
3.57 2.01 2.80 7.38 1.68 2.06 3.39 13.92 1.61 1.85 2.08 2.20 1.38 3.06 1.15 3.16 5.85 3.48
0.77 0.45 0.17 0.83 0.78 0.89 0.69 0.81 0.43 0.41 0.70 0.27 0.72 0.88 0.94 0.75 0.44 0.64
2.08 1.35 1.10 2.39 2.11 2.94 1.80 2.29 1.33 1.30 1.83 1.18 1.90 2.93 3.95 1.99 1.34 1.99
b
c
2d
Scatter correction: SNV + D and regression method: MPLS. Standard deviation. The standard error of calibration. Coefficient of determination in calibration. Standard error of cross-validation. 1 minus the ratio of unexplained variance to total variance.
of metabolic transformations. Thus, increasing the content of limiting and functional amino acids is very important for improving rapeseed meal’s nutritional value. It is necessary to understand the mechanism of gene expression on protein and amino acids content across environments. The genetic control of seed protein and amino acids content in rapeseed is complicated. Grami and Stefansson (1977) and Grami et al. (1997) found that protein trait in rapeseed is strongly affected by the environment, exhibiting a low heritability. However, the results of a study by Wang and Qiu (1990) showed that the broad- and narrow-sense heritability of protein content was as high as 0.748 and 0.572, respectively. Recently, some results found that protein and amino acid contents of rapeseed were simultaneously controlled by genetic effects of embryo, cytoplasm, and maternal plant, of which the maternal genetic effects were most important, followed by embryo and cytoplasmic genetic effects. The total narrow-sense heritability for protein content was 64.17%. Therefore, improvement of protein and amino acid contents of rapeseed would be more efficient when selection is based on maternal plants than that on single seeds. Selection for improving protein content can therefore be conducted in early generations (Ren et al., 2005; Wu et al., 2005; Variath et al., 2009). In addition, rapeseed protein and amino acid content could be changed by molecular techniques (Nesi et al., 2008). Modification of protein and amino acid contents in rapeseed is feasible because of large variation in amino acids such as lysine content. So, it is more necessary to construct a nondestructive, fast, reliable and robust method for screening intermediate lines in early generations. A small size of the standard error (SECV) does not clearly express the availability of an NIRS calibration for quality traits evaluation. SD/SECV is the ratio of the standard deviation (SD) of the amino acids content to the SECV, which was regarded as a criterion for assessing comprehensively the usefulness of NIRS predictions based on variation in the calibration set, processing of spectra data, and the control of errors from chemical measurements (Fontaine et al., 2001). If the SD/SECV ratio is high, NIRS prediction makes it easy to divide a given amount of samples into low, medium, and high contents of the amino acid. The corresponding equations could predict the contents of the amino acid with strong ability. Large variation in the calibration set is the prerequisite to achieve high SD/SECV. Small lab error in wet chemical analysis is also important and should be effectively controlled (Mentink et al., 2006). Studies by Fontaine et al. (2001, 2002) elucidated mechanism to successfully establish NIRS models. In present study, the prediction effects of most equations for relative contents of amino acids decreased due to lab errors multiplied. The SD/SECV ratios for all amino acid calibrations were shown in present project (Table 2). If this ratio exceeds a value of 3, the calibration equation is very meaningful and could be used to predict the amino acid content, values below 2 and above 1.5 have limited prediction effects, but the results may be accepted in a breeding program. In this study the (SD/SECV) ratio values of >3 were obtained for most calibration equations. Eleven different amino acids could be predicted with very good precision (R2 from 0.91 to 0.98, 1 − VR from 0.90 to 0.97 and SD/SECV with 3.13 to 5.90). The best results were derived for histidine, glycine and lysine with the SD/SECV of 5.90, 5.22, and 5.13, respectively. Their corresponding R2 were above 0.97. The equations for contents of alanine showed better agreement between reference value and spectra by NIRS with moderate determination coefficients (R2 = 0.84, 1 − VR = 0.82 and SD/SECV = 2.37). The calibration model for valine (R2 = 0.62, 1 − VR = 0.58 and SD/SECV = 1.54) might explain less variation, and predict the amino acid with less accuracy, but could be employed qualitatively in breeding. The NIRS prediction equations
118
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
for total amino acid content also showed high coefficient of determination and SD/SECV, and low SEC and SECV. Therefore, calibration equations were successfully developed for 16 different amino acids, except for model of valine content which gave a low coefficient of determination in the calibration. These calibration equations showed high accuracy in the prediction of amino acid contents of rapeseed meal. The results of NIRS calibration for amino acids in rapeseed meal were first reported with intact rapeseeds. When compared with the absolute contents, the relative contents of amino acids had poorer effects on the calibration equations. Each individual amino acid has lab error and the total AA has lab error too. When the relative content came from the individual absolute content divided by TAA, thus, it is lab error divided by error and the relative content error goes up × fold. So the relative contents all suffer from dual analytical error and lead to worse NIRS calibration effects. NIRS can predict amino acids contents in cereals much better than crude protein regressions, especially for lysine and methionine (Fontaine et al., 2001, 2002). In the present study, the result of correlation between individual amino acid content and TAA displayed the proportion of the variation explained merely by the TAA (Table 2). Most proportions between individual amino acids and TAA were smaller than R2 in calibration or 1 − VR in cross-validation for absolute contents. Additionally, all the relative contents except for serine and tyrosine show high R2 in calibration, whose the calibration effects of absolute content are associated with both individual amino acid and TAA (Table 3). Because of the bonds constitute, all NIR spectra of the amino acids were associated with the common bonds and specific bonds. The calibration of 14 individual amino acids contributed significantly, which suggested that their calibrations are promising, except for those for isoleucine, leucine and phenylalanine (R2 with 0.92–0.96) because isoleucine, leucine and phenylalanine highly correlate with TAA (R2 with 0.91–0.95). However considering the advantages of NIR technology, the calibrations on these three individual amino acid contents are still essential. The results indicated that the smaller the correlation coefficient between an individual amino acid content and TAA, the more reliable the calibration equation becomes. Calibration models are superior to traditional correlation coefficient method in prediction of content of amino acids in rapeseed meal. The NIRS analysis with intact rapeseeds offers an advantage of avoiding grinding or other sample preparations. The results of this study showed that the NIRS calibrations for amino acids could be directly used in breeding programs. In addition, one of the main advantages of this technique is that the content of amino acids are simultaneously analyzed along with the fatty acid content, and other traits, such as glucosinolate content and oil content of rapeseed. Therefore, these results demonstrated that NIRS is a powerful tool for nondestructive assessment of variation for amino acid contents in rapeseed. NIRS as a cost-effective and useful analytical tool has considerable potential in the rapeseeds improvement program for the identification and screening of large numbers of breeding materials in the early stage of breeding. 5. Conclusion Optimizing sample presentation and combination of data processing are very important for reliable and excellent equations. Combination of SNV + D/“2,4,4,1”/MPLS/was the best method in this study. Future research should aim at enhancing the accuracy of the prediction by adding new samples to the calibration set, reducing the lab error of the wet chemistry, and optimizing spectra data processing. Recalibration and further optimization of the calibration equations will probably yield an improvement in their performance, and make them more suitable on large scale in rapeseed breeding. The purpose of developing these NIRS equations was to obtain a rapid method for estimating individual and total amino acid contents based on intact rapeseeds. The results exhibited that NIRS can be used as a screening method to evaluate a large number of samples in a short period of time. The NIRS calibration equations developed in this study allow for simultaneous analysis along with other quality traits of rapeseed in a nondestructive, high-speed and cost-effective way. Acknowledgements This research was financially supported from Technology Office of Zhejiang Province (No. 2008C22084), Foundation for University Key Teacher by the Ministry of Education of China and by the 151 Program for the Talents of Zhejiang Province. We thank Mr. Yuanlong Li, the technician of 985-Institute of Agribiology and Environmental Sciences of Zhejiang University, for the assistance in using the L-8900 amino acid analyzer. References Abe, H., Kusama, T., Kawano, S., Iwamoto, M., 1995. Nondestructive determination of amino acid content in a single kernel of wheat and soybean by near infrared spectroscopy. In: Davies, A.M.C., Williams, P. (Eds.), Near Infrared Spectroscopy: The Future Waves. NIR Publications, Chichester, UK, pp. 457–461. AOAC, 1980. Official Methods of Analysis of the Association of Official Analytical Chemists, 13th edition. The Association, Washington, DC. Barnes, R.J., Dhanoa, M.S., Lister, S.J., 1989. Standard normal variant transformation and de-trending of near-infrared reflectance spectra. Appl. Spectrosc. 43, 772–777. Baye, T.M., Pearson, T.C., Settles, A.M., 2006. Development of a calibration to predict maize seed composition using single kernel by near infrared spectroscopy. J. Cereal. Sci. 43, 236–243. Ben-Gera, I., Norris, K.H., 1968. Determination of moisture content in soybeans by direct spectrophotometry. Israelic J. Agric. Res. 18, 125–132. Downey, R.K., Harvey, B.L., 1963. Methods of breeding for oil quality in rape. Can. J. Plant Sci. 43, 271–275. Fontaine, J., Hörr, J., Schirmer, B., 2001. Near-infrared reflectance spectroscopy enables the fast and accurate prediction of the essential amino acid contents in soy, rapeseed meal, sunflower meal, peas, fishmeal, meat meal products, and poultry meal. J. Agric. Food Chem. 49, 57–66.
G.L. Chen et al. / Animal Feed Science and Technology 165 (2011) 111–119
119
Fontaine, J., Schirmer, B., Hörr, J., 2002. Near-infrared reflectance spectroscopy (NIRS) enables the fast and accurate prediction of essential amino acid contents. 2. Results for wheat, barley, corn, triticale, wheat bran/middlings, rice bran, and sorghum. J. Agric. Food Chem. 50, 3902–3911. Goding, L.A., Downey, R.K., Finlayson, A.J., 1972. Seed protein amino acid compositions resulting from crosses between two Brassica campestris cultivars. Can. J. Plant Sci. 52, 63–71. Grami, B., Baker, R.J., Stefansson, B.R., 1997. Genetics of protein and oil content in summer rape: heritability, number of effective factors and correlations. Can. J. Plant Sci. 57, 937–943. Grami, B., Stefansson, B.R., 1977. Gene action for protein and oil content in summer rape. Can. J. Plant Sci. 57, 625–631. Hom, H.H., Heiko, C., Möllers, C., 2007. Non-destructive analysis of rapeseed quality by NIRS of small seed samples and single seeds. Euphytica 153, 27–34. Huisman, J., Tolman, G.H., 1990. Antinutritional factors in the plant proteins of diets for non-ruminants. In: Haresign, M., Cole, D.J.A. (Eds.), Recent Advances in Animal Nutrition. Butterworth–Heinemann, Oxford, pp. 3–31. Josefesson, E., Muhlenberg, C., 1968. Glucosinolates in seed of Swedish crops. Acta Agric. Scand. 18, 97–100. Mentink, R.L., Hoffman, P.C., Bauman, L.M., 2006. Utility of near-infrared reflectance spectroscopy to predict nutrient composition and in vitro digestibility of total mixed rations. J. Dairy Sci. 89, 2320–2326. Nesi, N., Delourme, R., Brégeon, M., Falentin, C., Renard, M., 2008. Genetic and molecular approaches to improve nutritional value of Brassica napus L. seed. CR Soc. Biol. 331, 763–771. Pujol, S., Pérez-Vendrell, A.M., Torrallardona, D., 2007. Evaluation of prediction of barley digestible nutrient content with near-infrared reflectance spectroscopy (NIRS). Livest. Sci. 109, 189–192. Ren, Y.L., Wu, J.G., Shi, C.H., 2005. Analysis of embryo, cytoplasmic and maternal effects on three amino acid traits in rapeseed. J. Zhejing Univ. (Agricultural and Life Science) 31, 41–46. Röbbelen, G., 1981. Potential and restrictions of breeding for amino acid improvement in rapeseed. In: Bunting, E.S. (Ed.), Production and Utilization of Amino acid in Oilseed Crops. Martinus Nijhoff, The Netherlands, pp. 3–11. Roussel, V., Branlard, G., Vézine, J.C., Bertrand, D., Balfourier, F., 2005. NIRS analysis reveals temporal trends in the chemical composition of French bread wheat accessions cultivated between 1800 and 2000. J. Cereal. Sci. 42, 193–203. Shenk, J.S., Westerhaus, M.O., 1991a. Population definition, sample selection, and calibration procedures for near infrared spectroscopy. Crop Sci. 31, 469–474. Shenk, J.S., Westerhaus, M.O., 1991b. Population structuring of near infrared spectra and modified partial least squares regression. Crop Sci. 31, 1548–1555. Shenk, J.S., Westerhaus, M.O., 1993. Analysis of Agriculture and Food Products by Near Infrared Reflectance Spectroscopy. Infrasoft International, Port Matilda, USA. Shenk, J.S., Westerhaus, M.O., 1995. Routine Operation, Calibration Development and Network System Management Manual for Near Infrared Instruments Version 3. 1. Infrasoft International, Port Matilda, USA. Simbaya, J., Alominski, S.B., Rakow, G., Downey, R.K., 1995. Quality characteristics of yellow-seeded Brassica seed meals: protein, carbohydrates, and dietary fiber components. J. Agric. Food Chem. 43, 2062–2066. Theander, O., Aman, P., Westlund, E., Andersson, R., Pettersson, D., 1995. Total dietary fiber determined as neutral sugar residues, uronic acid residues, and klason lignin (the Uppsala method): collaborative study. J. AOAC Int. 78, 1030–1044. Variath, M.T., Wu, J.G., Li, Y.X., Chen, G.L., Shi, C.H., 2009. Genetic analysis for oil and protein contents of rapeseed (Brassica napus L.) at different developmental times. Euphytica 166, 145–153. Velasco, L., Becker, H.C., 1998a. Analysis of total glucosinolate content and individual glucosinolates in Brassica spp. by near-infrared reflectance spectroscopy. Plant Breed. 117, 97–102. Velasco, L., Becker, H.C., 1998b. Estimating the fatty acid composition of the oil in intact-seed rapeseed (Brassica napus L.) by near-infrared reflectance spectroscopy. Euphytica 101, 221–230. Velasco, L., Fernández, J.M., De Haro, A., 1997. Determination of the fatty acid composition of the oil in intact-seed mustard by near-infrared reflectance spectroscopy. J. Am. Oil Chem. Soc. 74, 1595–1602. Velasco, L., Möllers, C., 2002. Nondestructive assessment of protein content in single seeds of rapeseed (Brassica napus L.) by near-infrared reflectance spectroscopy. Euphytica 123, 89–93. Velasco, L., Möllers, C., Becker, H.C., 1999. Estimation of seed weight, oil content and fatty acid composition in intact single seeds of rapeseed (Brassica napus L.) by near infrared reflectance spectroscopy. Euphytica 106, 79–85. Wang, F., Qiu, J., 1990. Studies on the inheritance of seed protein content and its correlation with other characters in Brassica napus L. Sci. Agric. Sin. 23, 42–47. Wu, G.Y., 2009. Amino acids: metabolism, functions, and nutrition. Amino Acids 37, 1–17. Wu, J.G., Shi, C.H., 2004. Prediction of grain weight, brown rice weight and amylose content in single rice grains using near-infrared reflectance spectroscopy. Field Crops Res. 87, 13–21. Wu, J.G., Shi, C.H., Zhang, H.Z., 2005. Genetic analysis of embryo, cytoplasmic, and maternal effects and their environment interactions for protein content in Brassica napus L. Aust. J. Agric. Res. 56, 69–73. Wu, J.G., Shi, C.H., Zhang, H.Z., Fan, L.J., 2002a. Study on analysis of oil content in intact rapeseed with few sample size by near infrared reflectance spectroscopy. Acta Agric. Sin. 28, 421–425. Wu, J.G., Shi, C.H., Zhang, X.M., 2002b. Estimating the amino acid composition in the milled rice powder by near-infrared reflectance spectroscopy. Field Crops Res. 75, 1–7.