Calibration and Prediction of Amino Acids in Stevia Leaf Powder

Nov 7, 2011 - In the present study, 301 samples of stevia leaf powder were defined as the calibration set from which calibration models were optimized...
0 downloads 0 Views 797KB Size
ARTICLE pubs.acs.org/JAFC

Calibration and Prediction of Amino Acids in Stevia Leaf Powder Using Near Infrared Reflectance Spectroscopy Guan Li,† Ruiguo Wang,† Alfred Julius Quampah,† Zhengqin Rong,† Chunhai Shi,*,† and Jianguo Wu*,†,‡ †

Department of Agronomy, College of Agriculture and Biotechnology, Zhejiang University, Hangzhou 310058, People’s Republic of China ‡ Department of Horticulture, College of Agriculture and Food Science, Zhejiang A and F University, Lin’an 311300, People’s Republic of China ABSTRACT: The use of stevia as animal feed additive has been researched over the years, but how to rapidly predict its amino acid contents has not been studied yet by using near-infrared reflectance spectroscopy. In the present study, 301 samples of stevia leaf powder were defined as the calibration set from which calibration models were optimized, and the performance of prediction was evaluated. Compared with other mathematical treatments, the models developed with the “1, 12, 12, 1” treatment, combined with modified partial least-squares regression and standard normal variance with de-trending, had a significant potential in predicting amino acid contents, such as threonine, serine, etc. Six spectral regions were found to possess large spectrum variation and show high contribution to calibration models. From the present study, the calibration models of amino acids in stevia were successfully developed and could be applied to quality control in feed processing, breeding selection and mutant screening. KEYWORDS: amino acid, near infrared reflectance spectroscopy (NIRS), stevia, animal feed additive, calibration, modified partial least squares regression (MPLS)

’ INTRODUCTION Stevia rebaudiana Bertoni, a member of the Compositae family, is a plant native to certain regions of South America (Paraguay and Brazil)1 and has been used as a natural sweetener for hundreds of years. The leaves of the stevia plant contain more than 8 different glycosides,2 including stevioside and rebaudoside A, which are the major constituents.3,4 Stevioside, the primary glycoside, is 250 to 300 times sweeter than sucrose.5 In recent years, stevia has been grown and applied as a natural sweetener in many countries such as Brazil, Paraguay, China, Japan, and in some parts of Europe and South America.6,7 Stevioside serves as a noncalorific stable dietary supplement for humans,8 which can also reduce the intake of sugar, and is useful for diabetic and phenylketonuria patients as well as obese persons.9 In addition, there are some patents, in which stevia has been used as an additive in animal feed. For example, Kim10 created a method which produced feed additives effectively using stevia, which had high antioxidant activity and no toxic effects. Doen11 also obtained livestock feed additives from stevia leaves and stalks. Without using synthetic agrochemicals, the feed additive from stevia is safe and free from drug induced pollution or damage. Stevia as a feed additive has been investigated over the years. Shiozaki et al.12 reported that extracts from stevia stalk might avert histamine toxicity in the rainbow trout stomach. Other researchers have also published their findings on the issue. Ghanta et al.13 found that stevia might be useful as a potential source of natural antioxidants, and Munro et al.14 indicated that there may be potential for stevia as a feed additive for piglets. However, the potential uses and practical implications of stevia in livestock and poultry diets are still a subject of research.15 Since stevia serves as a feed additive, its amino acid contents can affect the feed formulation and can consequently influence r 2011 American Chemical Society

the growth and development of livestock and poultry. Therefore, amino acid composition constitutes an important quality index for animal feed. The determination of amino acid contents in feed and its additive is very significant research and becomes as routine work in feed formulation processes. Amino acids can be determined chemically using amino acid autoanalyzer and high performance liquid chromatography (HPLC), with high precision. However, these standard methods are also too complex to manage, time-consuming, destructive, and expensive16,17 as well. Therefore, an alternative analytical method that is rapid and nondestructive could be a better option. Rubenthaler and Bruinsma18 first reported a successful calibration of lysine content in cereals using near-infrared reflectance spectroscopy (NIRS). This was followed by the prediction of amino acids in feed using NIRS by a number of authors.19 The major advantages of NIRS are reduced labor, lower timeconsumption, multiplicity of analyses, nonconsumption of samples, and low cost.20,21 Samples can be determined in various statuses, such as paste, powder, or even liquid.22 With NIRS, the energies emitted by chemical bonds, formed between hydrogen and other atoms such as C, O, N, and S, can be indirectly measured,23 as a result of which the contents of many substances in samples can be determined using NIRS.24 Amino acids generally possess at least two or more C-H, O-H, and N-H groups,25 making them suitable for NIRS determinations. NIRS can determine many nutritional contents in feed rapidly and precisely and thus is considered as a common powerful Received: September 6, 2011 Accepted: November 7, 2011 Revised: November 5, 2011 Published: November 07, 2011 13065

dx.doi.org/10.1021/jf2035912 | J. Agric. Food Chem. 2011, 59, 13065–13071

Journal of Agricultural and Food Chemistry analytical tool for quality control in feed mills.26,27 NIRS has been extensively recognized as a worthy and powerful analytical tool in the accurate determination of the chemical compositions28,29 in various samples. Recently, NIRS combined with pattern recognition techniques, has received much attention as it revealed its great ability for chemical content analysis and its application in many different fields.30,31 Furthermore, NIRS can make great contributions to crop breeding by serving as a guide in the selection of breeding materials. So far, no report has been found in relation to the use of NIRS for the quantitative detection of amino acids in stevia. The objective of the current study was therefore to examine the potential of NIRS to obtain accurate and robust calibrations, developed from the spectra of stevia leaf powder, to predict the contents of a wide range of amino acids that can be applied to the selection of stevia breeding materials.

ARTICLE

Table 1. Statistics of Amino Acid Contents (%) in Stevia Leaf Powder Samples for NIRS Analysis calibration constituent

’ MATERIALS AND METHODS Samples Preparation. A total of 520 strains of stevia were harvested at the farm of Zhejiang University. Before data acquisition, 5 10 branches from each line were cut during the budding stage and sun-dried for a few days. Then, the dried leaves were collected and ground into powder by using a grinder Model DFT-50 (Linda Technique Ltd., Zhejiang). Samples were thoroughly mixed in between grindings, and the particle size of the powder was controlled below 80 meshes. Powdered samples were placed in 12  8 cm plastic sealed bags and stored at 15 °C in a stockroom. Furthermore, the sealed bags of all the samples were opened and placed in a tightly closed oven for more than 48 h to balance the moisture at 8% before spectra scanning and chemical analyses. Chemical Analyses. About 100 mg of the powdered leaf samples were weighed to make sure that the amino acid concentration of the test solution would be similar to that of the standard mixture (Type H, Wako Pure chemical industrial Ltd., Japan). The samples were placed into hydrolysis tubes with the addition of 5 mL of HCl (6 mol/L). The tubes were then sealed with an alcohol burner after air was removed with a vacuum pump and placed in an oven for 24 h at 110 °C for protein hydrolysis. The samples were left to cool down, after which they were transferred to a 50 mL flask for dilution. A 5 mL solution was taken and evaporated to dryness by Rotary Evaporator (RE-52AA, Shanghai), and 10 mL 0.02 mol/L HCl was added to dissolve all the amino acids. Finally, about 1.5 mL of solution was filtered into a sample bottle through a 0.22 μm membrane for analysis. The amino acid autoanalyzer model L8900 (Hitachi, Japan) was used to determine different amino acids present in the samples. Amino acids were separated by a cation exchanger resin, postcolumn reacted with ninhydrin, and detected at 570 nm by a photometer. The amount of each amino acid present in each sample was calculated with reference to the standard mixed sample and expressed as a percentage by the software EZChrom Elite (Hitachi, Japan). NIRS Measurement. Samples containing about 2.5 g were loaded and pressed slightly to obtain similar packing density. The loading time was as short as possible to avoid excessive moisture absorption. All the samples were scanned by the NIRS monochromator, and the corresponding spectra were collected using a NIRSystems 5000 (Silver Spring, USA) instrument in reflectance mode. Thirty-two scans were performed for both the reference and each sample. Acquisition of the spectra was accomplished in the wavelength range from 1100 to 2500 nm with an interval of 2 nm by using the WinISI II (InfraSoft International, USA) software. In addition, each sample was loaded and scanned 4 times, and the average spectrum of each of the four recordings was used for NIR analysis.

range

external validation a

mean

SD

range

mean

SD

Asp

1.063 2.499

1.727

0.278 0.980 2.395

1.658

0.252

Thr

0.493 1.035

0.754

0.099 0.481 0.883

0.744

0.080

Ser Glu

0.546 2.773 1.294 2.534

1.027 1.905

0.339 0.538 1.816 0.237 1.233 2.368

1.036 1.894

0.295 0.199

Gly

0.588 1.107

0.856

0.100 0.542 1.009

0.846

0.084

Ala

0.623 1.291

0.954

0.115 0.570 1.235

0.941

0.101

Val

0.600 1.230

0.948

0.108 0.576 1.164

0.935

0.101

Ile

0.473 1.001

0.723

0.095 0.453 0.870

0.716

0.081

Leu

0.848 1.778

1.300

0.168 0.800 1.550

1.288

0.142

Tyr

0.278 0.732

0.495

0.076 0.320 0.602

0.485

0.062

Phe Lys

0.576 1.166 0.676 1.463

0.887 1.062

0.116 0.528 1.040 0.151 0.679 1.293

0.867 1.050

0.096 0.129

His

0.240 0.465

0.343

0.045 0.211 0.406

0.337

0.036

Arg

0.529 1.160

0.813

0.124 0.480 0.987

0.796

0.100

Pro

0.657 3.491

1.729

0.581 0.716 2.924

1.671

0.503

TAAb

11.033 20.699 15.523 1.877 10.467 18.263 15.265 1.423

a

SD: Standard deviation and the unit as %. b TAA: Total sum of all of the above amino acids of each sample. For the purpose of selecting a suitable sample set for calibration, the algorithms CENTER and SELECT were progressed, with the Mahalanobis distance (Global H distance, GH) under 3 and the neighborhood H distance (NH) under 0.9. Using the former algorithm, a population of 519 samples was defined after one sample was found to be an outlier. Using the latter algorithm, a total of 301 samples were selected from all the 519 stevia leaf powder samples. From the 301 samples, 225 samples were randomly selected for the calibration, and the remaining 76 were kept for the external validation. Spectral Data Processing. Since NIRS determinations are likely to be affected by the physical properties of the samples and other surrounding noises,32 it is essential to perform mathematical processing to heighten the contribution of the chemical composition to the spectral signal by reducing the systematic noise. In this study, some spectral preprocessing methods were applied comparatively. Spectra were first treated with “standard normal variance with de-trending (SNV-D)”, which is a method for scatter correction (recommended by WinISI to reduce particle size noise), and the first or second derivative of the calibration spectra was calculated with 3 gaps of data points. With regards to the 301 samples, up to 16 terms were allowed, though the software generally stopped much below the limit using the crossvalidation results as criterion, and for cross-validation, there were 5 or 6 groups. Results without scatter correction and mathematical treatments were regarded as the control. The regression algorithm named as modified partial least-squares regression (MPLS) was used for the calibration process. Considering the large amount of calculation involved, just the MPLS algorithm, based on the most important variations in the spectra, as well as the reference data, which transforms the spectra data points to terms,33 was applied in this experiment. The standard error of calibration (SEC), coefficient of calibration determination (RSQ), 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 models obtained and to determine the best calibration model, while the standard error of prediction (SEP), slope, and coefficient of determination in external validation (R2) were used to evaluate the external validation effects on 13066

dx.doi.org/10.1021/jf2035912 |J. Agric. Food Chem. 2011, 59, 13065–13071

Journal of Agricultural and Food Chemistry

ARTICLE

Table 2. Comparision of the Effects among the Calibration Equations of the Content of the Three Amino Acids with Different Mathematical Treatments calibration maths 0, 0, 1, 1

1, 8, 8, 1

1, 12, 12, 1

2, 8, 8, 1

2, 12, 12, 1

average

external validation c

d

e

SEC

RSQ

SECV

1 VR

SEP

slope

R2f

SD/SEP

Thr

0.029

0.916

0.031

0.903

0.030

0.896

0.875

2.674

Ile

0.027

0.921

0.028

0.913

0.032

0.914

0.846

2.522

His

0.015 0.024

0.888 0.908

0.016 0.025

0.873 0.896

0.016 0.026

0.940 0.917

0.815 0.845

2.269 2.488

Thr

0.029

0.915

0.031

0.900

0.031

0.886

0.876

2.588

Ile

0.025

0.928

0.027

0.913

0.031

0.903

0.864

2.603

His

0.014

0.901

0.015

0.876

0.016

0.948

0.817

2.269

Thr

0.029

0.914

0.031

0.900

0.030

0.890

0.878

2.674

Ile

0.025

0.926

0.027

0.914

0.030

0.908

0.867

2.690

His

0.013

0.916

0.015

0.890

0.015

0.899

0.842

2.420

Thr Ile

0.030 0.025

0.911 0.925

0.032 0.028

0.898 0.909

0.029 0.030

0.901 0.939

0.881 0.868

2.767 2.690

His

0.013

0.909

0.015

0.890

0.015

0.932

0.841

2.420

0.023

0.916

0.025

0.899

0.025

0.912

0.859

2.569

0.030

0.902

0.032

0.889

0.031

0.902

0.864

2.588 2.690

average 2, 4, 4, 1

b

constituent

average 1, 4, 4, 1

a

Thr Ile

0.025

0.930

0.027

0.916

0.030

0.930

0.868

His

0.015

0.885

0.017

0.856

0.016

0.974

0.811

2.269

Thr

0.031

0.901

0.032

0.893

0.030

0.903

0.874

2.674

Ile His

0.027 0.013

0.916 0.910

0.028 0.016

0.908 0.874

0.030 0.015

0.924 0.944

0.868 0.840

2.690 2.420

Thr

0.030

0.904

0.031

0.897

0.030

0.902

0.879

2.674

Ile

0.027

0.918

0.028

0.910

0.030

0.917

0.866

2.690

His

0.014

0.904

0.015

0.878

0.015

0.952

0.834

2.420

0.024

0.908

0.025

0.891

0.025

0.928

0.856

2.568

a

SEC: the standard error of calibration and the unit as %. b RSQ: coefficient of determination in calibration. c SECV: standard error of cross-validation and the unit as %. d 1 VR: 1 minus the ratio of unexplained variance to total variance. e SEP: the standard error of prediction and the unit as %. f 2 R : coefficient of determination in external validation. the calibration models.34 The ratio of standard deviation (SD) with the standard error of prediction (SEP) (SD/SEP) was introduced to evaluate the external validation results between the chemical components when calibration models were developed.

’ RESULTS Chemical Analysis. Table 1 shows an overview of the selected amino acid contents in the stevia leaves. The statistics were expressed as the range, mean, and standard deviation (SD), determined by the chemical method. The average contents of glutamic acid (Glu, 1.905%), aspartic acid (Asp, 1.727%), leucine (Leu, 1.300%), proline (Pro, 1.729%), serine (Ser, 1.027%), and lysine (Lys, 1.062%) were much higher than that of other amino acids including threonine (Thr), glycine (Gly), isoleucine (Ile), alanine (Ala), valine (Val), phenylalanine (Phe), histisine (His), tyrosine (Tyr), and arginine (Arg). The contents of amino acids were different within each sample, and there were also significant differences between these 3 parameters among amino acids. The ranges and means of the amino acids in validation set were similar to those in the calibration set, and their SDs in the validation set were a little lower than that in the calibration set. Since there was an obvious variation in the contents of 15 amino acids in the calibration sample set, it was appropriate and representative for NIRS calibration. The content of total amino acids in validation was essentially identical with that in calibration.

Mathematical Treatments. Using MPLS regressions, with the first and second derivative combined with SNV-D, 6 calibrations were conducted for 3 main amino acids using the population data of the 301 stevia samples, and another one without any mathematical treatment as the control. The performance statistics are shown in Table 2. When compared with different mathematical treatments, the calibration effects were significantly different for these 3 amino acids. In general, the calibration effects between the first and second derivative were similar based on their average effects, and both were better than that of the control. In the first derivative treatments, the “1, 12, 12, 1” treatment displayed lower SECV and SEP, higher 1 VR, R2, and SD/SEP. Meanwhile, the second derivative treatments (“2, 8, 8, 1” and “2, 12, 12, 1”) showed similar effects and much better results compared to the “2, 4, 4, 1” treatment based on the performance in the external validation. When compared to the control, these 3 treatments (“1, 12, 12, 1”, “2, 8, 8, 1”, and “2, 12, 12, 1”) showed better effects on calibration and external validation. According to their SD/SEP values, the above 3 treatments showed similar validation effects on Ile (2.690) and His (2.420), but for Thr, the best treatment was “1, 12, 12, 1” with the highest value of 2.767. Therefore, the “1, 12, 12, 1” treatment was utilized to develop calibration for all 15 amino acids in the present experiment. Amino Acid Calibration Models. On the basis of the calibration results in Table 2, the pretreatment “1, 12, 12, 1” combined 13067

dx.doi.org/10.1021/jf2035912 |J. Agric. Food Chem. 2011, 59, 13065–13071

Journal of Agricultural and Food Chemistry

ARTICLE

Table 3. Statistic Results of NIRS Equations of Amino Acid Contents with Mathematical Treatments (1,12,12,1) and Scatter Corrections (SNV-D)a calibration

a

external validation

constituent

SEC

RSQ

SECV

1 VR

SEP

slope

R2

SD/SEP

Asp

0.070

0.935

0.083

0.908

0.098

0.894

0.779

2.572

Thr

0.029

0.915

0.030

0.906

0.031

0.891

0.878

2.588

Ser Glu

0.102 0.078

0.911 0.891

0.119 0.089

0.877 0.858

0.116 0.086

0.957 0.986

0.844 0.812

2.540 2.317

Gly

0.027

0.925

0.028

0.919

0.028

0.935

0.896

3.010

Ala

0.044

0.850

0.049

0.813

0.048

0.963

0.770

2.099

Val

0.039

0.871

0.041

0.858

0.048

0.945

0.779

2.104

Ile

0.025

0.928

0.026

0.921

0.029

0.931

0.874

2.782

Leu

0.042

0.937

0.044

0.931

0.050

0.923

0.886

2.839

Tyr

0.040

0.722

0.041

0.706

0.035

0.884

0.700

1.763

Phe Lys

0.036 0.042

0.905 0.922

0.039 0.048

0.886 0.900

0.042 0.048

0.875 0.979

0.834 0.867

2.282 2.693

His

0.014

0.901

0.015

0.886

0.015

0.950

0.830

2.420

Arg

0.052

0.821

0.058

0.777

0.058

0.848

0.685

1.717

Pro

0.141

0.939

0.160

0.921

0.191

0.918

0.861

2.635

TAA

0.510

0.926

0.549

0.914

0.570

0.932

0.849

2.496

For abbreviations of SEC, RSQ , SECV, 1 VR, SEP, and R2, see Table 2.

Figure 1. First derivative spectra (1, 12, 12, 1; SNV-D) from raw spectra of the stevia calibration samples.

with SNV-D was used to develop the calibration equations for all amino acids, and the statistics are shown in Table 3. The equations for 13 amino acids (Asp, Thr, Ser, Gly, Ile, Leu, Lys, Pro, Glu, Ala, Val, Phe, and His) presented high coefficients of determination for calibration (RSQ: 0.850 0.939) and crossvalidation (1 VR: 0.813 0.931). In addition, SD/SEP radios were more than 2.0 for these amino acids, and the slope values of the equations were close to 1. This indicated that a reliable and accurate estimation of these amino acids could be obtained by using NIRS. However, due to the low RSQ (0.722 and 0.821)

and the SD/SEP ratios (1.763 and 1.717) being less than 2.0 for Tyr and Arg, these calibration equations may be unsuitable for use in routine analysis. Since the calibration effect for the total sum of all amino acids (TAA) was also accurate, it also demonstrated that it was possible to estimate different amino acids with calibration equations which were developed from a unique calibration set. Spectra Analysis. The first derivative spectra (“1, 12, 12, 1”; SNV-D) from raw spectra of all the stevia calibration samples are shown in Figure 1. The peaks were very clear and sharp. It can be 13068

dx.doi.org/10.1021/jf2035912 |J. Agric. Food Chem. 2011, 59, 13065–13071

Journal of Agricultural and Food Chemistry

ARTICLE

Figure 2. Correlation plot of the total sum of all amino acid reference values and wavelength absorbance by using mathematic treatment (1, 12, 12, 1) and scatter correlation (SNV-D).

observed that the spectra significantly varied at 6 spectral ranges of 1386 1440 nm, 1660 1712 nm, 1880 1920 nm, 1946 2192 nm, 2234 2300 nm, and 2410 2452 nm, which included a lot of information related to the chemical components in the stevia leaf. The correlation plot in Figure 2 showed the result of correlation analysis for the total sum of all amino acid reference values and wavelength absorbance obtained from using the “1, 12, 12, 1” mathematical treatment and scatter correction (SNV-D). There was a significant positive correlation between the reference value and spectra absorbance at 2052, 2252, and 2452 nm, as well as an obvious negative correlation between them at 1446, 1916, and 2134 nm. The correlation coefficients were 0.638, 0.496, 0.796, 0.718, 0.614, and 0.866. Thus, these spectral regions showed larger contributions to the calibration model for amino acids.

’ DISCUSSION The present research was focused on amino acid calibration and prediction using NIRS, and the results indicated that this method could provide robust calibration models for most amino acids, although for now, it could not yield an ideal result for Tyr and Arg. The inaccuracy for the prediction of these two amino acids was due to the low 1 VR and high SECV of their respective calibration models (Table 3). In order to improve the prediction accuracy for these two amino acids, the calibration models should be redeveloped by enlarging the calibration population. Therefore, the samples with higher contents of these amino acids should be found by screening the germplasm or creating the mutants in future studies. From the results in Tables 2 and 3, the average effect of the first derivative and second derivative was significant, and the performances were both better than that without any mathematical treatment. However, the second derivative did not show a significant improvement over the first derivative, probably due to over fitting. Figures 1 and 2 indicated that there were significant variations or highly positive correlations among the spectra in the range between 2036 and 2066 nm. This might be mainly because the characteristic spectra absorption peak of

amide existed around 2050 nm, and the characteristic absorption peak of protein was around 2060 nm. Because of the primary overtone of the O-H group in water or some other components and a combination of stretching around 1440 nm, there were vast differences in absorption between 1386 and 1440 nm. The complexity of the chemical groups and components resulted in the absorbance of every wavelength point being caused by several types of bending, so it was impossible to find a spectrum data point which was strongly connected with a substance. Tryptophan was completely destroyed during hydrolysis, while methionine and cystine were partially destroyed; thus, no accurate data for the three amino acids could be obtained by the method of hydrochloric acid hydrolysis. To study these three amino acids, other hydrolysis methods and detection by UV spectrophotometer or some other special rapid methods could be used. Fontaine et al.33,35 have tried to predict methionine and cystine using NIRS by transferring them to methionine sulfone and cysteic acid at first; however, the effect of calibration and prediction were still not as good as those of other amino acids. Also, there must be an amount of free amino acids in the stevia leaf, but compared to the contents of amino acid from protein, they are very low. Carpena-Ruiz et al.36 extracted free amino acids from tomato leaves, and their research indicated that free amino acid contents in the leaf were very low, just tens of micrograms per gram fresh leaf. Also, to determine the contents of free amino acids, another detection method by an amino acid autoanalyzer model L-8900, which is called the biological fluids method, or some other measuring equipment such as HPLC and GC-MS could be used. Without the process of hydrolysis, there are other factors containing pigment and polyphenol, etc., which affect the accurate and precise determination of amino acid contents. Lin et al.31 proved that free amino acid content in Radix Pseudostellariae can be analyzed by NIRS; however, it is yet a significant but difficult work to accurately predict each of all of the free amino acids. Some researchers have proved with accurate results that NIRS could be used for amino acid prediction in animal feed.35 But few papers have reported that the NIRS had been successful in predicting amino acids in feed additives. The present study is 13069

dx.doi.org/10.1021/jf2035912 |J. Agric. Food Chem. 2011, 59, 13065–13071

Journal of Agricultural and Food Chemistry the first to report that NIRS can be used to predict the contents of amino acids in the new feed additive stevia. The overall results showed that NIRS coupled with the MPLS algorithm permits the simultaneous analysis of amino acid content in stevia. In this study, the NIRS analysis of stevia provided accurate predictions for Asp, Thr, Ser, Gly, Ile, Leu, Lys, and Pro. More research is required to confirm these results, improve the model performance of Glu, Ala, Val, Phe, and His, and to successfully predict Tyr and Arg. This will help reduce cost and save time. This technique could be introduced in the quality control of additives in feed processing and in the selection of early generations of stevia for breeding programs and in mutation library construction.

’ AUTHOR INFORMATION Corresponding Author

*(C.S.) Tel/Fax: +86-571-88982691. E-mail: [email protected]. (J.G.W.) Tel: +86-13516816401. E-mail: [email protected]. Funding Sources

This work has been financially supported by the Technology Office Project of Zhejiang Province (No. 2009C32030) and the 151 Program for the Talents of Zhejiang Province.

’ ACKNOWLEDGMENT We are grateful to 985 Institute of Agrobiology and Environmental Science of Zhejiang University and also thank Qiongqiong Zhu and Bin Zhang for their expert technical guidance. ’ ABBREVIATIONS USED NIRS, near-infrared reflectance spectroscopy; PCR, principal component regression; PLS, partial least-squares regression; MPLS, modified partial least-squares; SD, standard deviation; SEC, the standard error of calibration; SECV, standard error of crossvalidation; SEP, standard error of prediction; SD/SECV, ratio of the standard deviation (SD) of the amino acid content in the calibration samples to the standard error (SECV) of NIRS; SNV-D, standard normal variance with de-trending; RSQ, coefficient of determination in calibration; 1 VR, one minus the ratio of unexplained variance to total variance; R2, coefficient of determination in calibration; TAA, total of amino acid ’ REFERENCES (1) Hracek, V. M.; Gliemmo, M. F.; Campos, C. A. Effect of steviosides and system composition on stability and antimicrobial action of sorbates in acidified model aqueous systems. Food Res. Int. 2010, 43, 2171–2175. (2) Brandle, J. E.; Telmer, P. G. Steviol glycoside biosynthesis. Phytochemistry 2007, 68, 1855–1863. (3) Dacome, A. S.; Silva, C. C.; da Costa, C. E.M.; Fontana, J. D.; Adelmann, J.; da Costa, S. C. Sweet diterpenic glycosides balance of a new cultivar of Stevia rebaudiana (Bert.) Bertoni: Isolation and quantitative distribution by chromatographic, spectroscopic, and electrophoretic methods. Process Biochem. 2005, 40, 3587–3594. (4) Williams, L.; George, D.; Burdock, A. Genotoxicity studies on a high-purity rebaudioside A preparation. Food Chem. Toxicol. 2009, 47, 1831–1836. (5) Brusick, D. J. A critical review of the genetic toxicity of steviol and steviol glycosides. Food Chem. Toxicol. 2008, 46, S83–S91. (6) Geuns, J. M. C.; Augustijns, P.; Mols, R.; Buyse, J. G.; Driessen, B. Metabolism of stevioside in pigs and intestinal absorption characteristics

ARTICLE

of stevioside, rebaudioside A and steviol. Food Chem. Toxicol. 2003, 41, 1599–1607. (7) Geuns, J. M. C.; Malheiros, R. D.; Moraes, V. M. B.; Decuypere, E. M. P.; Compernolle, F.; Buyse, J. G. Metabolism of stevioside by chickens. J. Agric. Food Chem. 2003, 51, 1095–1101. (8) Hearn, L. K.; Subedi, P. P. Determining levels of steviol glycosides in the leaves of Stevia rebaudiana by near infrared reflectance spectroscopy. J. Food Compos. Anal. 2009, 22, 165–168. (9) Geuns, J. M. C. Molecules of interest stevioside. Phytochemistry 2003, 64, 913–921. (10) Kim, Y. G. Preparing feed-stuff additive using dioxin decomposer from Stevia, involves fermenting aqueous mixture of Stevia using yeast, low temperature drying and sterilizing fermented mixture, and mixing dried mixture with tuna fish meal powder. Patent KR2008109350-A. (11) Doen, F. Livestock feed additive for use in feed of livestock e.g. cow and pig, contains fermented and aged concentrated liquid extract of mixture of dry powder of ripened stevia stalk and leaves, further mixed with mixture. Patent JP2005124536-A. (12) Shiozaki, K.; Nakano, T.; Yamaguchi, T.; Sato, M.; Sato, N. The protective effect of stevia extract on the gastric mucosa of rainbow trout Oncorhynchus mykiss (Walbaum) fed dietary histamine. Aquacult. Res. 2004, 35, 1421–1428. (13) Ghanta, S.; Banerjee, A.; Poddar, A.; Chattopadhyay, S. Oxidative DNA damage preventive activity and antioxidant potential of Stevia rebaudiana (Bertoni) Bertoni, a natural sweetener. J. Agric. Food Chem. 2007, 55, 10962–10967. (14) Munro, P. J.; Lirette, A.; Anderson, D. M.; Ju, H. Y. Effects of a new sweetener, stevia, on performance of newly weaned pigs. Can. J. Anim. Sci. 2000, 80, 529–531. (15) Atteh, J. O.; Onagbesan, O. M.; Tona, K.; Decuypere, E.; Geuns, J. M. C.; Buyse, J. Evaluation of supplementary stevia (Stevia rebaudiana, bertoni) leaves and stevioside in broiler diets: effects on feed intake, nutrient metabolism, blood parameters and growth performance. J. Anim. Physiol. Anim. Nutr. 2008, 92, 640–649.  lvarez-García, N.; Gonzalez-Cabrera, (16) Gonzalez-Martín, I.; A J. M. Near-infrared spectroscopy with a fibre-optic probe for the prediction of the amino acid composition in animal feeds. Talanta 2006, 69, 706–710. (17) Wu, J. G.; Shi, C. H.; Zhang, X. M. Estimating the amino acid composition in milled rice by near-infrared reflectance spectroscopy. Field Crops Res. 2002, 75, 1–7. (18) Rubenthaler, G. L.; Bruinsma, B. L. Lysine estimation in cereals by near infrared reflectance. Crop Sci. 1978, 18, 1039–1042. (19) Sales, J. The use of linear regression to predict digestible protein and available amino acid contents of feed ingredients and diets for fish. Aquaculture 2008, 278, 128–142. (20) Skeie, S.; Feten, G.; Almøy, T.; Østlie, H.; Isaksson, T. The use of near infrared spectroscopy to predict selected free amino acids during cheese ripening. Int. Dairy J. 2006, 16, 236–242. (21) Wu, J. G.; Shi, C. H. Prediction of grain weight, brown rice weight and amylose content in single rice grains using near-infrared reflectance spectroscopy. Field Crops Res. 2004, 87, 13–21. (22) Stuart B. Infrared Spectroscopy: Fundamentals and Applications; John Wiley & Sons: Chichester, England, 2004; pp 1 3. (23) Park, R. S.; Agnew, R. E.; Gordon, F. J.; Steen, R. W. J. The use of near infrared reflectance spectroscopy (NIRS) on undried samples of grass silage to predict chemical composition and digestibility parameters. Anim. Feed Sci Technol. 1998, 72, 155–167. (24) Shen, F.; Niu, X. Y.; Yang, D. T.; Ying, Y. B.; Li, B. B.; Zhu, G. Q.; Wu, J. Determination of amino acid in Chinese rice wine by fourier transform near-infrared spectroscopy. J. Agric. Food Chem. 2010, 58, 9809–9816. (25) Heigl, N.; Huck, C. W.; Rainer, M.; Najam-ul-Haq, M.; Bonn, G. K. Near infrared spectroscopy, cluster and multivariate analysis hyphenated to thin layer chromatography for the analysis of amino acids. Amino Acids 2006, 31, 45–53. (26) Van Kempen, T.; Bodin, J. C. Near-infrared reflectance spectroscopy (NIRS) appears to be superior to nitrogen-based regression as 13070

dx.doi.org/10.1021/jf2035912 |J. Agric. Food Chem. 2011, 59, 13065–13071

Journal of Agricultural and Food Chemistry

ARTICLE

a rapid tool in predicting the poultry digestible amino acid content of commonly used feedstuffs. Anim. Feed Sci. Technol. 1998, 76, 139–147. (27) Pujol, S.; Perez-Vendrell, A. M.; Torrallardona, D. Evaluation of prediction of barley digestible nutrient content with near-infrared reflectance spectroscopy (NIRS). Livest. Sci. 2007, 109, 189–192. (28) Shenk, J. S.; Westerhaus, M. O. Accuracy of NIRS instruments to analyze forage and grain. Crop Sci. 1985, 25, 1120–1122. (29) Redshaw, E. S.; Mathision, G. W.; Milligan, L. P.; Weisenburger, R. D. Near-Infrared reflectance spectroscopy for prediction forage composition and voluntary consumption and digestibility in cattle and sheep. Can. J. Anim. Sci. 1986, 66, 103–115. (30) Ravn, C.; Skibsted, E.; Rasmus, B. Near-infrared chemical imaging on pharmaceutical solid dosage forms-comparing common calibration approaches. J. Pharm. Biomed. Anal. 2008, 48, 554–561. (31) Lin, H.; Chen, Q. S.; Zhao, J. W.; Zhou, P. Determination of free amino acid content in Radix Pseudostellariae using near infrared (NIR) spectroscopy and different multivariate calibrations. J. Pharm. Biomed. Anal. 2009, 50, 803–808. (32) Wu, J. G.; Shi, C. H. Calibration model optimization for rice cooking characteristics by near infrared reflectance spectroscopy (NIRS). Food Chem. 2007, 103, 1054–1061. (33) Fontaine, J.; Schirmer, B.; H€orr, J. 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. 2002, 50, 3902–3911. (34) Shenk, J. S.; Westerhaus, M. O. Population structuring of near infrared spectra and modified partial least squares regression. Crop Sci. 1991, 31, 1548–1555. (35) Fontaine, J.; H€orr, J.; Schirmer, B. Near-infrared reflectance spectroscopy enables the fast and accurate prediction of the essential amino acid contents in soy, rapeseed meal, sunflower meal, peas, fish meal, meat meal products and poultry meal. J. Agric. Food Chem. 2001, 49, 57–66. (36) Carpena-Ruiz, R.; Sope~na, A.; Ramon, A. M. Extraction of free amino acids from tomato leaves. Plant Soil 1989, 119, 251–254.

’ NOTE ADDED AFTER ASAP PUBLICATION There was an error in the R2 column heading of Table 2 in the version of this paper published November 18, 2011. The correct version published November 22, 2011.

13071

dx.doi.org/10.1021/jf2035912 |J. Agric. Food Chem. 2011, 59, 13065–13071