Evaluation of Non-Starch Polysaccharides and Oligosaccharide Content of Different Soybean Varieties ( Glycine Max ) by near Infrared Spectroscopy

Full text

Evaluation of Nonstarch Polysaccharides and Oligosaccharide
Content of Different Soybean Varieties (
Glycine max
)by
Near-Infrared Spectroscopy and Proteomics
KRISTIN HOLLUNG,*,† MARGARETH ØVERLAND,‡MILICA HRUSTICÄ,§
PETAR SEKULICÄ,§JEGOR MILADINOVICÄ,§HARALD MARTENS,†BJØRG NARUM,†
STEFAN SAHLSTRØM,†METTE SØRENSEN,‡TROND STOREBAKKEN,‡AND
ANDERS SKREDE‡
Matforsk, The Norwegian Food Research Institute, Osloveien 1, N-1430 Ås, Norway, Aquaculture
Protein Centre, Centre of Excellence, Post Office Box 5003, N-1430 Ås, Norway, and Institute of
Field and Vegetable Crops, Maksima Gorkog 30, 21000 Novi Sad, Serbia and Montenegro
A total of 832 samples of soybeans were screened by near-infrared (NIR) reflectance spectroscopy,
to identify soybean samples with a lower content of oligosaccharides and nonstarch polysaccharides
(NSP). Of these, 38 samples were identified on the basis of variation in protein content and agronomic
value and submitted to high-resolution NIR spectroscopy. On the basis of the NIR data, 12 samples
were further selected for chromatographic characterization of carbohydrate composition (mono-, di-,
and oligosaccharides and NSP). Their soluble proteins were separated by two-dimensional gel
electrophoresis (2DE). Using partial least-squares regression (PLSR), it was possible to predict the
content of total NSP from the high-resolution NIR spectra, suggesting that NIR is a suitable and
rapid nondestructive method to determine carbohydrate composition in soybeans. The 2DE analyses
showed varying intensities of several proteins, including the glycinin G1 precursor. PLSR analysis
showed a negative correlation between this protein and insoluble NSP and total uronic acid (UA).
KEYWORDS: Near-infrared spectroscopy; soybean;
Glycine max
; carbohydrate composition; proteomics;
nonstarch polysaccharides
INTRODUCTION
Soybean meal is a major source of protein in diets for
monogastrics throughout the world, but its use for certain species
is limited by the presence of a wide variety of antinutritional
factors (1,2). Reduced growth performance, increased viscosity
of digesta, reduced nutrient digestibility (3,4,5), increased
incidence of diarrhea (6), and undesirable morphological
changes of the intestinal epithelium (7,8,9) have been observed
when soybean meal has been used in substantial amounts in
feed for monogastrics. While the growth-depressing effect of
heat-labile antinutritional factors in soybeans can be mostly
overcome by processing, heat-stable antinutritional factors, such
as nonstarch polysaccharides (NSP) and oligosaccharides, are
not eliminated (1,10). Raw soybean meal contains about 30%
carbohydrates: approximately 20% NSP and 10% oligosaccha-
rides (11). Soybean NSP consist mainly of arabinans, arabinoga-
lactans, and acidic polysaccharides; approximately one-third of
the NSP are soluble (12). The primary oligosaccharides in
soybeans are the galacto-oligosaccharides, raffinose, stachyose,
and verbascose (13). Both oligosaccharides and NSPs are known
to cause digestive disorders and reduced performance in
monogastrics (3,7,14). Identifying soybean varieties with low
levels of NSP and oligosaccharides would increase the potential
for soybeans as an alternative protein source for monogastrics.
Near-infrared (NIR) spectroscopy is a tool to predict chemical
composition of various feed ingredients. Rapid assessment of
feed ingredients enables accurate feed formulation and can be
used by plant breeders to select new cultivars. While NIR
spectroscopy has traditionally been used to predict gross
chemical content of feed ingredients (15), limited information
exists on the use of NIR spectroscopy to predict carbohydrate
composition. To our knowledge, no published papers address
the efficacy of NIR to predict the content of complex carbo-
hydrates in soybeans.
Proteomics is a powerful tool allowing identification of
proteins involved in the determination of functionality of an
organism (16,17). Proteomics of mature seeds provide more
general information of the proteome than proteomics of
metabolizing organisms such as growing plants because protein
turnover is lower in mature seeds. Recently, a proteome
reference map for soybean was published, which will be useful
* To whom correspondence should be addressed: Matforsk, Osloveien
1, N-1430 A°s, Norway. Telephone: +47-64-97-01-42. Fax: +47-64-97-
03-33. E-mail: kristin.hollung@matforsk.no.
†The Norwegian Food Research Institute.
‡Aquaculture Protein Centre.
§Institute of Field and Vegetable Crops.
9112
J. Agric. Food Chem.
2005,
53,
9112
−
9121
10.1021/jf051438r CCC: $30.25 © 2005 American Chemical Society
Published on Web 10/14/2005

for identification of proteins from 2DE (18). In this study,
expression of 679 proteins from developing soybean seeds were
profiled, and of these, 422 proteins were identified. Two other
studies of mature soybean seeds have identified 44 and 100
proteins, respectively (19,20). Comparative proteomics may
be used as a tool for the selection of soybean varieties with
specific nutritional properties.
A study was carried out to (1) evaluate different soybean
varieties by the use of NIR and proteomics with respect to the
content of protein and different carbohydrates to identify suitable
alternative protein sources for monogastrics, (2) evaluate the
use of NIR as a rapid screening method to predict content of
complex carbohydrates of soybeans, and (3) determine whether
specific seed proteins are correlated to the amount of specific
NSP and oligosaccharides in soybeans.
MATERIALS AND METHODS
Selection of the Soybean Genotypes. The soybean genotypes were
obtained by the Institute of Field and Vegetable Crops, Novi Sad, Serbia
and Montenegro. A total of 832 genotypes of soybeans were initially
screened by NIR spectroscopy in the 950-1700 nm region (DA 7000
FLEXI-MODE NIR/vis spectrophotometer, Perten Instruments, Hud-
dinge, Sweden). The screening was performed on whole soybeans by
transmittance measurements. On the basis of predictions of chemical
data on protein and carbohydrate composition, 30 genotypes were
chosen for analysis by high-resolution NIR spectroscopy. Samples 1-30
were grown in the experimental fields in the Novi Sad area of
Vojvodina, Serbia and Montenegro (45°16′N and 19°51′E) in 2004.
The dominant soil type at the field where the genotypes were grown is
the calcareous chernozem with a pH of 7.65 and organic matter content
of 3.3%. Samples 1-8 and 31-38 are the same genotypes, but these
samples were obtained from various test plots in the Vojvodina region
in 2004 (21). There were thus 38 soy samples in total. The predictions
of chemical composition are shown in Table 1. Ground soybean samples
were made by milling 2 ×100 g samples in an IKA Universalmu¨hle
(IKA Works, Wilmington, NC) for 15 s in liquid nitrogen. The predicted
crude protein values of the 38 soybean samples were confirmed by
chemical analysis (Kjeldahl N ×6.25) (EU Dir. 93/28).
NIR Spectroscopy of Soybeans. For the 38 samples, high-resolution
reflectance spectra from 400 to 2498 nm was measured in a Foss
NIRSystems model 6500 scanning monochromator (Silver Spring, MD)
equipped with a transport module (2 nm steps, 32 scans per sample).
Absorbance values were recorded as log 1/R, where Ris the sample
reflectance. The analysis of unground and ground samples was carried
out using a coarse granular transport cell. This cell is rectangular with
internal dimensions of 4.1-cm wide, 17.2-cm long, and 1.4-cm deep.
When the ground samples were analyzed, the instrument was turned
90°(placed on a stand with the coarse cell in the horizontal direction),
not to permit any of the sample to sprinkle out of the cell. Ceramic
was used as a reference. Each sample was placed into the cell and
scanned twice. These data formed the basis for selecting 12 particularly
informative samples for detailed chemical analysis.
Carbohydrate Analysis. For the 12 selected samples, total and
insoluble neutral NSP contents and their sugar compositions were
determined according to the method described by Englyst et al. (22),
by measuring the neutral sugars in acidic hydrolysates by gas
chromatography (GC) as alditol acetates. The soluble fraction was
calculated by the difference. Uronic acid (UA) was analyzed spectro-
photometrically (22).
Table 1.
Predicted Content of Moisture, Crude Protein, Fat, Carbohydrates, and Ash in the 38 Soybean Samples (% Wet Weight)
sample genotype variety moisture crude protein crude fat crude carbohydrates ash
1 NS-L-115 JELICA 7.7 40.7 23.7 23.2 4.8
2 NS-L-66 KRAJINA 8.3 42.4 21.5 22.9 5.0
3 NS-L-200143 FORTUNA 8.7 44.0 19.7 22.5 5.1
4 NS-L-400007 MELI 8.7 42.2 21.2 23.2 4.7
5 NS-L-2023 AFRODITA 8.4 40.6 21.9 24.5 4.7
6 NS-L-101095 PROTEINKA 8.2 41.2 22.5 23.2 5.0
7 NS-L-201149 VALJEVKA 8.3 39.3 22.3 25.1 5.0
8 NS-L-101136 LARA 8.9 39.8 21.7 24.0 5.6
9 NS-L-201167 ALISA 8.8 41.5 21.9 23.0 4.8
10 NS-L-201187 BEC
ˇ
EJKA 8.4 41.5 21.4 23.9 4.9
11 NS-L-401004 TARA 8.4 41.1 21.9 24.0 4.7
12 NS-L-401009 8.5 40.2 22.2 24.2 4.8
13 NS-L-110 BALKAN 8.8 40.3 21.9 24.1 5.0
14 NS-L-2022 RAVINICA 6.8 39.5 21.8 27.0 4.9
15 NS-L-2107 NOVOSA
\
ANKA 7.6 42.0 21.5 24.1 4.8
16 NS-L-110146 ANA 7.5 39.9 22.2 25.8 4.6
17 NS-L-110168 8.4 38.5 23.8 24.6 4.6
18 NS-L-110175 8.0 40.5 21.8 25.1 4.7
19 NS-L-110181 TEA 8.2 40.3 23.1 23.8 4.6
20 NS-L-110190 8.1 40.0 22.8 24.7 4.5
21 NS-L-210174 ZVEZDA 7.4 39.3 20.9 27.7 4.8
22 NS-L-210200 SAVA 8.0 40.0 21.9 25.2 4.9
23 NS-L-210201 S
ˇ
APC
ˇ
ANKA 7.8 39.7 21.4 26.0 5.2
24 NS-L-2024 VOJVO
\
ANKA 6.7 39.1 21.9 27.5 4.9
25 NS-L-220203 10.3 39.2 21.1 24.6 4.9
26 NS-L-220207 10.8 40.3 20.4 23.5 5.1
27 NS-L-120169 MIMA 7.2 37.2 24.0 26.9 4.8
28 NS-L-330219 MORAVA 7.8 38.9 23.0 25.4 4.9
29 NS-L-110120 VENERA 7.5 38.6 23.0 26.2 4.8
30 NS-L-430005 11.5 39.4 21.1 23.0 5.1
31 NS-L-115 JELICA 10.1 50.1 18.2 18.4 5.1
32 NS-L-66 KRAJINA 10.8 47.3 19.0 19.5 5.1
33 NS-L-200143 FORTUNA 11.2 49.9 17.2 19.9 5.2
34 NS-L-400007 MELI 11.6 48.2 16.9 21.1 5.3
35 NS-L-2023 AFRODITA 9.7 50.4 19.0 19.3 4.9
36 NS-L-101095 PROTEINKA 7.4 36.6 26.4 25.2 4.3
37 NS-L-201149 VALJEVKA 8.4 37.2 24.5 28.1 4.4
38 NS-L-101136 LARA 7.8 38.9 24.3 28.7 4.3
Evaluation of NSP and Oligosaccharide Content of Soybeans
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 9113

The concentration of low-molecular-weight sugars was quantified
using an HPLC technique based on that of Copp et al. (23) with a few
modifications. A total of5gofdried and milled sample was shaken
with 25 mL of 50% methanol with 1.5 mg mL-1mannitol as an internal
standard for 30 min. Activated carbon (2.5 g) was added, and the
suspension was shaken for 60 min at room temperature and left at 4
°C for 60 min. The samples were filtrated by paper filters. Filtrate was
collected; 5 mL of the filtrates was incubated at 35 °C for 16 h; and
precipitates were removed by centrifugation. A total of 1 mL of filtrate
was evaporated by vacuum centrifugation (ISS110 SpeedVac, Termo
Savant) and redissolved in 1 mL of distilled water. After filtration
through Millex-HV (0.45 µm, 13 mm), the samples were analyzed by
HPLC on a Shimadzu pump (LC-10AD) controlled by Class VP
software. The sample (20 µL) was injected with a SIL-10 autoinjector
into a Varian Carbohydrate PB-column (300 ×7.8 mm) eluted with
water (0.4 mL/min) at 80 °C and equipped by a refractive detector
Figure 1.
High-resolution NIR spectroscopy of the 38 soybean samples. Spectra were collected from 400 to 2498 nm and expressed as absorbance.
The upper 38 spectra are log(1/
T
) from whole soybeans, and the lower 38 spectra are log(1/
R
) from ground soybeans. The rectangle marks the region
from 2000 to 2450 nm known to contain information on the carbohydrate composition.
Figure 2.
NIR spectra of the 38 soybean samples from the carbohydrate-relevant region 2000
−
2450 nm after MSC. Spectra for whole soybeans are
shown to the left, and spectra from ground soybeans are shown to the right.
9114
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 Hollung et al.

(RID-6A). The sugars were quantified on the basis of their areas relative
to the internal standard and corrected for their individual response
factors. All samples were analyzed in triplicates.
Protein Extraction and 2DE. For the 12 samples, a total protein
extract was prepared according to the protocol in Weiss et al. (24). All
chemicals were analytical-grade from either Merck (Whitehouse Station,
NJ), Sigma-Aldrich (St. Louis, MO), or BioRad Laboratories (Her-
cules, CA). Briefly, 50 mg of ground soybeans was extracted in 500
µL of buffer [6 M urea, 1% Triton-X-100, 0.5% DTT, and 0.5% (w/v)
IPG buffer pH 3-10 (Amersham Biosciences, Piscataway, NJ)]. Protein
content was measured using the RC-DC protein assay (BioRad). The
total protein extract (50-500 µg) was mixed with rehydration buffer
(8 M urea, 2 M thiourea, 2% Nonidet P-40, 2% IPG buffer pH 3-10,
and 60 mM DTT) at a total of 350 µL and used for in-gel rehydration
of homemade 18-cm immobilized pH-gradient (IPG) strips pH 4-9.
Isoelectric focusing was performed using a stepwise protocol for 35
kVh using the IPGphor apparatus (Amersham Biosciences). IPG strips
were equilibrated in buffer containing 50 mM Tris-HCl at pH 8.8, 6
M urea, 30% glycerol, 2% SDS, and 10 mg/mL DTT for 15 min and
then for 15 min in equilibration buffer containing 25 mg/mL
Figure 3.
PLSR of NIR spectra from the 38 soybean samples. NIR spectra from both whole and ground soybeans are used as
x
variables, and samples
are used as
y
variables. Predictions of protein, ash, moisture, fat, and carbohydrate contents are used as guidance in the analysis.
Table 2.
Total Neutral NSP Content, Uronic Acid Content, and Content of the Individual Monosaccharides of Soybean Samples as (g kg
-
1
) Dry
Matter Basis
a
sample rhamnose fucose arabinose xylose mannose galaktose glucose total NSP total uronic acid
4 11.3 a
b
10.3 a 36.5 c 31.1 cd 13.2 b 60.1 bc 107.1 bcd 269.6 de 24.7 abc
5 9.1 a 4.5 a 42.8 b 43.1 a 21.7 a 72.6 abc 125.5 a 319.3 a 26.5 a
14 8.9 a 5.3 a 46.0 ab 38.3 ab 17.8 ab 74.1 ab 119.7 ab 310.1 ab 25.2 ab
17 8.9 a 5.9 a 46.9 ab 31.2 cd 17.3 ab 73.4 abc 111.5 bc 295.1 bc 26.2 ab
28 8.2 a 5.3 a 43.2 b 28.0 de 16.0 ab 78.0 a 91.4 ef 270.1 de 21.9 bc
29 8.6 a 5.6 a 43.1 b 31.2 cd 16.7 ab 76.5 a 89.5 ef 271.2 de 23.1 abc
10 9.0 a 5.7 a 45.6 ab 31.4 cd 17.4 ab 77.7 a 104.7 cd 291.5 bc 25.5 ab
26 9.3 a 6.5 a 47.4 ab 34.9 bc 16.1 ab 81.4 a 95.5 de 291.1 bc 23.5 abc
32 8.8 a 5.0 a 43.6 b 28.2 de 13.7 b 66.7 abc 90.7 ef 256.7 e 23.4 abc
35 6.9 a 4.3 a 35.3 c 23.5 e 14.4 b 57.4 c 78.7 f 220.5 f 20.4 c
36 8.2 a 5.4 a 49.8 a 34.6 bc 18.3 ab 80.2 a 88.4 ef 284.9 cd 25.0 ab
38 10.2 a 5.8 a 49.8 a 34.0 bc 16.2 ab 78.1 a 99.2 cde 293.3 bc 24.5 abc
p
value 0.128 0.485 <0.001 <0.001 0.01 0.001 <0.001 <0.001 0.005
a
Data are means of two replicates.
b
Means followed by a different letter within the same column are significantly different (
p
< 0.05).
Table 3.
Insoluble Neutral NSP Content, Uronic Acid Content, and Content of the Individual Monosaccharides of Soybean Samples as (g kg
-
1
) Dry
Matter Basis
a
sample rhamnose fucose arabinose xylose mannose galactose glucose insoluble NSP insoluble uronic acid
4 6.8 a
b
4.0 bc 31.5 cd 32.4 abc 11.1 ab 51.3 bc 78.1 de 215.2 de 23.4 ab
5 6.9 a 4.1 abc 36.2 b 34.8 ab 13.3 a 56.6 ab 108.4 a 260.3 a 25.9 a
14 6.9 a 4.4 ab 38.1 ab 35.8 a 12.2 ab 58.7 ab 99.3 ab 255.4 ab 24.3 ab
17 6.6 a 4.8 ab 40.6 a 28.1 bcde 11.3 ab 57.4 ab 89.9 bc 238.7 bc 26.1 a
28 5.7 a 4.1 abc 34.7 bc 24.0 de 11.0 ab 59.6 ab 70.0 ef 209.1 de 24.4 ab
29 5.7 a 4.5 ab 34.9 bc 26.4 cde 11.5 ab 60.0 ab 69.3 ef 212.3 de 24.7 ab
10 6.7 a 4.7 ab 36.5 b 22.9 e 11.6 ab 60.9 ab 82.9 cd 226.2 cd 23.2 ab
26 5.4 a 4.9 a 36.4 b 25.4 cde 9.8 ab 60.3 ab 68.0 ef 210.2 ab 24.9 ab
32 6.5 a 4.0 bc 34.9 bc 21.6 e 10.7 ab 52.7 ab 71.1 ef 201.5 e 26.6 a
35 5.3 a 3.5 c 28.8 d 23.0 e 9.8 ab 42.1 c 63.4 f 175.9 f 20.6 b
36 5.9 a 4.7 ab 40.8 a 25.5 cde 9.7 ab 62.1 a 76.6 de 225.3 cd 25.3 a
38 5.4 a 4.2 abc 38.3 ab 30.2 abcd 10.1 b 55.2 ab 70.7 ef 214.1 de 25.5 a
p
value 0.05 <0.001 <0.001 <0.001 0.04 <0.001 <0.001 <0.001 0.009
a
Data are means of two replicates.
b
Means followed by a different letter within the same column are significantly different (
p
< 0.05).
Evaluation of NSP and Oligosaccharide Content of Soybeans
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 9115

iodoacetamide instead of DTT. After equilibration, proteins were
separated on vertical 12.5% polyacrylamide gels on the Ettan DALT
twelVeapparatus (Amersham Biosciences). Analytical gels were silver-
stained according to the protocol of Blum et al. (25). For preparative
gels, 500 µg of protein was used in the IEF and the gels were silver-
stained, but glycine was replaced by 5% acetic acid after development.
Gel alignments were performed with the Z3 3.0 program package from
Compugen, Inc. (San Jose, CA).
Protein Identification by Mass Spectrometry. Gel plugs containing
the protein spot of interest were washed twice for 15 min in 50 mM
ammonium carbonate at pH 8.0-8.5/acetonitril (1:1) and dried in a
speed-vac centrifuge for 20 min. A total of 30 µL of 50 mM ammonium
carbonate at pH 8.0-8.5 containing 0.15 µg of trypsin (Sequence grade,
Promega, Madison, WI) was added to each gel plug and incubated for
45 min on ice before transfer to 37 °C and further incubation overnight.
The supernatant was removed, added 3 µL of formic acid, and diluted
to a final concentration of 0.25% TFA. The sample was enriched and
purified using OMIX microcolumns (Varian, Palo Alto, CA) according
to instructions from the manufacturer. The peptide samples were eluted
in R-cyano-4-hydroxy-cinnamic acid, 0.1% TFA/50% ACN (1:1), and
spotted directly onto the target plate. Peptide samples were analyzed
by mass spectrometry (MS) using matrix-assisted laser desorption
ionization time-of-flight (MALDI-TOF) in an Ultraflex MALDITOF/
TOF (Bruker Daltronics, Billerica, MA). A peptide standard (Bruker
Daltronics) was used for external calibration of the instrument, and
internal calibration was performed using the trypsin autolysis peaks.
Proteins were identified by peptide mass fingerprinting, searching the
NCBI database using the MASCOT search engine at http://
www.matrixscience.com, and 100 ppm tolerance. All identified proteins
were confirmed by MS/MS using the LIFT module on 1-2 selected
parent ions and repeated database searches.
Statistics. The data were analyzed by ANOVA using the GLM
procedure of SAS (SAS version 6.08, SAS Institute, Inc., Cary, NC).
Tukey test (p<0.05) was performed to determine minimum significant
differences (MSD). Principal component analysis (PCA) and the
multivariate analysis partial least-squares regression (PLSR) were
carried out using The Unscrambler version 9.05 (CAMO ASA,
Trondheim, Norway). Because the outlined region from 2000 to 2450
nm of the NIR spectra is known to be particularly informative about
carbohydrate composition (26), this region was used for the multiplica-
tive signal correction (MSC).
RESULTS
High-Resolution NIR Spectroscopy of Soybeans. To in-
vestigate how to get maximum information about soybean
samples at minimum laboratory work, two different ways of
using NIR was tested. High-resolution NIR absorbance spectra
Table 4.
Content of Low-Molecular-Weight Carbohydrates in 12 Selected Soybean Varieties as (mg g
-
1
) Dry Matter Basis
a
sample stachyose raffinose sucrose maltose glucose xylose fructose
4 20.20 c
b
7.68 bcde 50.31 abc 1.95 cd 1.77 cde 9.18 bc 0.29 a
5 28.60 a 6.74 cde 47.74 abcd 3.29 ab 2.91 a 9.55 b 0.43 a
14 24.09 abc 9.15 ab 48.02 abcd 1.30 d 1.59 e 8.67 bcd 0.48 a
17 21.09 bc 8.60 bc 52.80 ab 2.30 bcd 2.40 abcde 12.87 a 0.63 a
28 25.30 ab 5.92 e 40.96 def 3.32 ab 2.29 abcde 8.85 bc 0.66 a
29 27.11 a 6.68 cde 44.00 cde 3.72 a 2.78 ab 9.25 bc 0.66 a
10 24.75 abc 7.95 bcd 55.01 a 1.75 cd 1.69 de 6.48 e 0.50 a
26 28.46 a 6.86 cde 45.77 bcd 2.63 abc 2.64 abc 7.00 de 0.37 a
32 27.30 a 9.51 ab 37.57 ef 2.02 cd 1.92 bcde 7.68 cde 0.70 a
35 24.61 abc 11.05 a 33.44 f 1.82 cd 1.81 cde 8.63 bcd 0.52 a
36 26.44 a 6.41 de 48.29 abcd 3.55 a 2.27 abcde 12.16 a 0.57 a
38 24.51 abc 7.07 cde 49.79 abc 3.60 a 2.58 abcd 13.28 a 0.30 a
p
value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.4824
a
Data are means of three replicates.
b
Means followed by a different letter within the same column are significantly different (
p
< 0.05).
Figure 4.
Loading plot of the two first dimensions of the PLSR of carbohydrate composition of the 12 selected soybean samples. Chromatographically
determined carbohydrates were used as
x
variables, and sample identifiers were used as
y
variables. Predictions of crude protein and carbohydrates are
included as passive variables in the figure.
9116
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 Hollung et al.

from 400 to 2500 nm of both whole and ground soybeans are
shown in Figure 1. Clear absorbances are evident in the visible
wavelength range. Characteristic water absorbance peaks are
visible around 1200, 1450, and 1940 nm, and around 2100 nm
the characteristic carbohydrate peak is visible, especially in the
upper, whole soybean curves. In contrast, the ground soybean
spectra show small but relatively sharp lipid peaks around 1720
and 2350 nm. (The more complex peak patterns of the soybean
proteins, known to contribute to the observed spectra, are more
difficult to identify by visual inspection alone.) Thus, the two
ways of measuring the soybeans gave somewhat different NIR
information, judging already from the visual appearance of the
curves. However, because of the extreme precision of the NIR
measurements, detailed multivariate mathematical feature ex-
traction is normally more informative than just visual inspection.
The obvious cross baseline and scaling variations in the spectra
represent light-scattering variations, which may be caused by
differences in the physical structure in the samples and thus to
differences in the behavior upon grinding. To separate the
physical information from the chemical information of the
samples, the spectra were corrected using MSC. Figure 2 shows
the effect of spectral preprocessing by MSC in the 2000-2450
nm region of the NIR spectra, intended to separate the two main
types of variation in the spectra in Figure 1, physical light
scattering and chemical light absorption. The MSC was
performed on NIR spectra in the 2000-2400 nm region for both
the whole and ground soybean samples. Figure 2 shows
differences in the spectra from whole and ground soybean
samples. Although most of the light-scattering variations are
removed, differences can still be observed both in general
baseline and scaling levels as well as in specific chemical
absorption peaks. To select a subset of the soybean samples
for further analyses, a PLSR model y≈f(x) was generated from
the information in the MSC-treated spectra (xvariables).
Predictions of protein, ash, and carbohydrate contents based on
initial NIR spectroscopic measures were used as yvariables for
modeling guidance in the selection model. This multivariate
calibration model demonstrated that the soybean samples had
different chemical composition. PLSR is a bilinear regression
modeling method, where a few latent variables (“PLS compo-
nents, PCs”) T)[t1,t2,t3, ...] are generated as y-relevant linear
combinations of the xvariables, T≈f1(x), and yis then modeled
by regression on these latent variables, y≈f2(T). Figure 3
represents the bilinear ”score plots” of t1versus t2, showing the
position of each of the 38 soybean samples with respect to the
two most important NIR variation patterns. This plot and the
corresponding score plot of t1versus t3was used for selecting
12 soybean samples for further investigation. These 12 samples
were chosen to form 6 different pairs (connected by line
segments in Figure 3). Samples 5 and 35 are the same genotype
but grown at different locations, even though they appear in
separate groups based on the NIR spectroscopic analysis and
measured chemical content. The pair containing samples 10 and
26 was chosen based on the originally predicted contents of
carbohydrates and protein. The predicted protein content was
highly correlated to the analyzed content (r)0.96).
Composition of Carbohydrates. The content of different
nonstarch carbohydrates was determined chromatographically
in the selected 12 samples. Tables 2 and 3show the content of
total and insoluble NSPs and UA, respectively. There were
significant differences among the soybean samples for the
content of all measured total monosaccharides, except for
rhamnose and fucose, and for the content of all measured
insoluble monosaccharides, except for rhamnose. The UA
content varied significantly among the samples, and most of
the UA was found in the insoluble NSP fraction. The oligosac-
charide composition of the 12 different soybean samples is
presented in Table 4. Among the raffinose group of oligosac-
charides, stachyose was predominant in these soybean samples.
With the exception of fructose, there were significant differences
among the 12 soybean samples for the content of low-molecular-
weight carbohydrates. A bilinear PLSR model was developed,
using carbohydrates as xvariables and sample indicator variables
(0/1) as noninformative yvariables. Additional relevant variables
are included for interpretation: crude protein and carbohydrate
contents predicted from the initial NIR analyses are included
as passified (down-weighted and thus noninfluential) yvariables.
The variation in the carbohydrate content was mainly explained
by the first two components. This model is summarized as the
scale-free correlation loading plot for the first two components
in Figure 4, where the axes represent the correlation coefficient
Figure 5.
Pairwise raw-data plots of selected carbohydrates for verification
of the PLSR results. (A) Correlation of maltose and raffinose content. (B)
Correlation of total UA and insoluble NSP. (C) Correlation of sucrose
and total UA.
Evaluation of NSP and Oligosaccharide Content of Soybeans
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 9117

between each of the input xand yvariables and PLSR score t1
(abscissa) and t1(ordinate), respectively. Positions are given
one character to the left of the variable names. The figure shows
that within the obtained model, total UA, total NSP, insoluble
NSP, sucrose, and crude carbohydrates are strongly positively
intercorrelated and negatively intercorrelated to the crude protein
measurement, which are particularly high in soybean sample
35. Maltose and raffinose are likewise negatively intercorrelated.
Stacchyose shows a strong, unique variation pattern and is
particularly high in sample 29 and low in sample 4. The most
salient features in these raw data are plotted in more detail in
Figure 5, which shows the best correlation among individual
carbohydrates.As seen in Figure 4,Figure 5 shows that there
is a negative correlation between raffinose and maltose (r)
-0.77) and a positive correlation between total UA and both
insoluble NSP (r)0.87) and sucrose (r)0.85).
NIR as a Predictor of Carbohydrate Composition. Using
the MSC-treated NIR spectra as xvariables and the carbohydrate
measurements as yvariables, a PLSR model was made. Cross-
validation between the 12 samples indicated 2 PLS PCs to give
the best predictive ability for all of the carbohydrate variables,
explaining on the average 42% of the 13 yvariables. Figure 6
shows the correlation loading plot for the yvariables for the
first two PCs. Because different variation patterns might affect
the different carbohydrate variables, individual PLSR calibration
models were made for each individual carbohydrate variable
(y) versus the MSC-treated spectra (x). Some of the best
prediction abilities of individual carbohydrates from NIR spectra
are shown in Figure 7. Full cross-validation was used for finding
the optimal number of PCs in each case. There was a positive
correlation between the NIR spectra and total NSP (r)0.91),
insoluble NSP (r)0.80), total UA (r)0.86), and raffinose (r
)0.75).
Changes in Protein Expression Related to Carbohydrate
Composition. The selected 12 samples were analyzed by 2DE
(Figure 8). Approximately 1000 proteins were resolved on the
gels, and 590 protein spots were matched in the data set. Several
proteins appeared to vary substantially in intensities among the
different samples, although the cross-validation did not reveal
any significant PLS PCs, probably because the number of
samples was too low compared to the heterogeneity of the
sample set. The protein marked by an arrow in the figure was
identified as the glycinin G1 precursor protein from soybean
(Swissprot ID P04776) with 30% sequence coverage. The
observed and theoretical pI and MW are in agreement (pI 5.9;
MW, 6.3 kD), and the gel position also matches the position in
the reference maps described in Hajduch et al. (18). A PLSR
model (Figure 9) of the spot intensities of the glycinin G1
precursor demonstrates that the intensity of the protein correlates
to some of the carbohydrates. There was a negative correlation
between the glycinine G1 precursor and both insoluble NSP (r
)-0.70) and total UA (r)-0.80) (Figure 10).
DISCUSSION
NIR spectroscopy was used in combination with traditional
chemical analyses to predict protein and carbohydrate composi-
tion in soybeans. The results of the present study suggest that
NIR spectroscopy is a promising tool to predict the content of
complex carbohydrates such as total and insoluble NSP, total
and insoluble UA, and some low-molecular-weight carbohy-
drates. NIR spectroscopy might thus be used as a basis to select
soybeans with desirable carbohydrate composition for defined
nutritional purposes. Changes in protein expression were found
to be associated with certain carbohydrate components in
soybeans, suggesting that protein composition is related to the
composition of carbohydrates.
A variation in the content of NSP and total UA among the
38 soybean samples was observed. Glucose comprised the
largest proportion of the total and insoluble NSP, followed by
galactose, arabinose, xylose, mannose, rhamnose, and xylose.
These results are similar to results presented earlier (27). Sucrose
Figure 6.
Correlation loading plot of carbohydrates modeled by NIR in the selected 12 soybean samples. NIR spectra of ground soybeans are used as
x
variables, and compositions of carbohydrates are used as
y
variables.
9118
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 Hollung et al.

comprised the largest proportion of the oligosaccharides, fol-
lowed by raffinose, stachyose, xylose, maltose, and glucose.
This is in agreement with earlier studies reporting that the main
oligosaccharides in soybean meal are the R-galactosides sucrose
(6-7%), raffinose (1-2%), and stachyose (5-6%), accounting
for a total soluble carbohydrate content of 12-15% (1,28).
Although soybean meal is a major protein supplement for
monogastric animals throughout the world, the oligosaccharides
and NSP from soybean have been shown to cause digestive
disorders and reduced performance in several species (3,7,14).
A negative effect of soybean NSP on the digestibility of fat
and protein has been reported in broiler chickens and salmon
(7). Oligosaccharides are not digested by enzymatic hydrolyses
in monogastrics because of the lack of the enzyme R-galac-
tosidase, which is necessary to hydrolyze the R-1,6 linkages
present in oligosaccharides (29), but are fermented by the
bacterial population in the gastrointestinal tract. Fermentation
of oligosaccharides in the gastrointestinal tract has been
associated with adverse effects on nutrient digestibility and
energy availability of soybean meal in poultry (30,31). In pigs,
the effect of oligosaccharides on nutrient digestibility is not as
clear (6,32), but reduced nutrient digestibility (4) and increased
diarrhea (6) have been reported. In fish, indigestible soybean
oligosaccharides have been reported to cause osmotic diarrhea
and alterations in the intestinal flora (33).
Despite the limited number of soybean samples analyzed for
complex carbohydrate composition, it was possible to obtain a
reasonably good model to predict total NSP, unsoluble NSP,
raffinose, and total UA content from the high-resolution NIR
spectra of the soybean samples. For most low-molecular-weight
carbohydrates, it was not possible to make a prediction model
based on the NIR spectra because of the large variation in the
content of these carbohydrates in the samples. NIR calibration
normally requires many more calibration samples to optimize
and assess the statistical calibration model. In particular, the
cross-validation was negatively impacted by sample 35, which
was rather unique, both in the NIR spectrum and in the
carbohydrate composition, compared to sample 32, which was
within the same group as sample 35 in the original selection of
the 12 samples. Hence, improved results for the NIR-based
prediction of the patterns of covariation among the carbohydrate
fractions may be attainable, but this needs further validation
and refinement based on more data.
Composition of soybean carbohydrates may be influenced
by genotype and environmental conditions, including soil type,
Figure 7.
Correlation plots between some NIR-predicted and chromato-
graphically measured carbohydrates in the selected 12 soybean samples.
(A) Prediction of total NSP. (B) Prediction of insoluble NSP. (C) Prediction
of total uronic acid. (D) Prediction of raffinose.
Figure 8.
Representative 2DE image of soybean. A total protein extract
of soybean was separated on IPG at pH 4
−
9 in the first dimension and
10% SDS
−
PAGE in the second dimension. Proteins are silver-stained,
and the spot identified as the glycinin G1 precursor is marked by an
arrow.
Evaluation of NSP and Oligosaccharide Content of Soybeans
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 9119

fertilizer application, and climate (13,21). In the present study,
large variation in the chemical content among the soybean
samples was observed among and within genotypes. For
instance, soybean samples 5 and 35 were from the same
genotype but grown under different environmental conditions.
The protein content in these two samples differed by 10%, with
the total protein and relative amount of glycinin G1 precursor
being higher in sample 35. In addition, the total NSP differed
by 10%, insoluble NSP differed by 8.4%, and the content of
raffinose was twice as high in sample 35 as compared with
sample 5. This indicates that the environmental conditions were
more important for the variation in the content of protein and
NSP than the effect of the genotype. However, because the
soybeans are grown in the field, it cannot be excluded that cross-
fertilization influenced the chemical composition.
In this study, we observed a negative correlation between
the seed storage protein G1 glycinin precursor and insoluble
NSP and raffinose. Glycinin and β-conglycinin have shown to
produce immunological reactions in young farm animals.
β-conglycinin appears to be more immunoreactice than glycinin
in young calves (9). To our knowledge, a correlation between
carbohydrate composition and seed storage proteins has not been
previously reported. Proteomics could be a powerful tool to
reveal more information on the proteome in soybeans related
to the composition of carbohydrates. Recently, an oilseed
proteomics initiative has published proteome maps with expres-
sion profiles of 679 protein spots during seed filling in soybeans
(18). These maps may be used as reference tools to identify
soybean seed proteins.
We have demonstrated that high-resolution NIR has a
potential to identify heat-stable antinutritional factors, such as
NSP and oligosaccharides in soybeans. This can aid in the
selection of suitable soybeans for use in diets for target
monogastric species.
ACKNOWLEDGMENT
We thank Anne Helene Bjerke for skillful technical assistance.
The equipment at the MS/Proteomics platform in the Food
Science Alliance, Ås, Norway, was used for identification of
proteins.
Figure 9.
Loading plot of the two first dimensions of the PLSR of carbohydrate composition and glycinin G1 precursor content in the selected 12
soybean samples. Carbohydrate compositions are used as
x
variables, and contents of the glycinin G1 precursor are used as
y
variables.
Figure 10.
Pairwise raw-data plots of the glycinin G1 precursor content
and selected carbohydrates for verification of the PLSR results. (A)
Correlation of glycinin G1 content and insoluble NSP. (B) Correlation of
glycinin G1 content and total UA.
9120
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 Hollung et al.

LITERATURE CITED
(1) Francis, G.; Makkar, H. P. S.; Becker, K. Antinutritional factors
present in plant-derived alternate fish feed ingredients and their
effects in fish. Aquaculture 2001,199, 197-227.
(2) Anderson, R. L.; Wolf, W. J. Compositional changes in trypsin
inhibitors, phytic acid, saponins, and isoflavones related to
soybean processing. J. Nutr. 1995,125, S581-S588.
(3) Smits, C. H. M.; Annison, G. Non-starch plant polysaccharides
in broiler nutritionsTowards a physiologically valid approach
to their determination. Worlds Poult. Sci. J. 1996,52, 203-221.
(4) Kornegay, E. T. Feeding value and digestibility of soybean hulls
for swine. J. Anim. Sci. 1978,47, 1272-1280.
(5) Skrede, A.; Krogdahl, A. Heat affects nutritional characteristics
of soybean meal and excretion of proteinases in mink and chicks.
Nutr. Rep. Int. 1985,32, 479-489.
(6) Zhang, L. Y.; Li, D. F.; Qiao, S. Y.; Johnson, E. W.; Li, B. Y.;
Thacker, P. A.; Han, I. K. Effects of stachyose on performance,
diarrhoea incidence, and intestinal bacteria in weanling pigs.
Arch. Anim. Nutr.-ArchiV. Fur Tierernahrung 2003,57,1-10.
(7) Refstie, S.; Svihus, B.; Shearer, K. D.; Storebakken, T. Nutrient
digestibility in Atlantic salmon and broiler chickens related to
viscosity and non-starch polysaccharide content in different
soyabean products. Anim. Feed Sci. Technol. 1999,79, 331-
345.
(8) Krogdahl, A.; Bakke-McKellep, A. M.; Baeverfjord, G. Effects
of graded levels of standard soybean meal on intestinal structure,
mucosal enzyme activities, and pancreatic response in Atlantic
salmon (Salmo salar L.). Aquacult. Nutr. 2003,9, 361-371.
(9) Lalle`s Soy Products as Protein Sources for Preruminants and
Young Pigs; Drackley, J. K., Ed.; Federation of Animal Science
Societies: Savoy, IL, 2000; pp 106-126.
(10) Leske, K. L.; Jevne, C. J.; Coon, C. N. Effect of oligosaccharide
additions on nitrogen-corrected true metabolizable energy of soy
protein concentrate. Poult. Sci. 1993,72, 664-668.
(11) Snyder, H. E.; Kwon, T. W. Soybean Utilization; Van Nostrand
Reinhold: New York, 1987; p 60.
(12) Knudsen, K. E. B. Carbohydrate and lignin contents of plant
materials used in animal feeding. Anim. Feed Sci. Technol. 1997,
67, 319-338.
(13) Karr-Lilienthal, L. K.; Kadzere, C. T.; Grieshop, C. M.; Fahey,
G. C., Jr. Chemical and nutritional properties of soybean
carbohydrates as related to nonruminants: A review. LiVestock
Prod. Sci. 2005, in press.
(14) Coon, C. N.; Leske, K. L.; Akavanichan, O.; Cheng, T. K. Effect
of oligosaccharide-free soybean meal on true metabolizable
energy and fiber digestion in adult roosters. Poult. Sci. 1990,
69, 787-793.
(15) Perez-Marin, D. C.; Garrido-Varo, A.; Guerrero-Ginel, J. E.;
Gomez-Cabrera, A. Near-infrared reflectance spectroscopy (NIRS)
for the mandatory labelling of compound feedingstuffs: Chemi-
cal composition and open-declaration. Anim. Feed Sci. Technol.
2004,116, 333-349.
(16) Hunter, T. C.; Andon, N. L.; Koller, A.; Yates, J. R.; Haynes,
P. A. The functional proteomics toolbox: Methods and applica-
tions. J. Chromatogr., B: Biomed. Sci. Appl. 2002,782, 165-
181.
(17) Tyers, M.; Mann, M. From genomics to proteomics. Nature 2003,
422, 193-197.
(18) Hajduch, M.; Ganapathy, A.; Stein, J. W.; Thelen, J. J. A
systematic proteomic study of seed filling in soybean. Establish-
ment of high-resolution two-dimensional reference maps, expres-
sion profiles, and an interactive proteome database. Plant Physiol.
2005,137, 1397-1419.
(19) Mooney, B. P.; Krishnan, H. B.; Thelen, J. J. High-throughput
peptide mass fingerprinting of soybean seed proteins: Automated
workflow and utility of UniGene expressed sequence tag
databases for protein identification. Phytochemistry 2004,65,
1733-1744.
(20) Herman, E. M.; Helm, R. M.; Jung, R.; Kinney, A. J. Genetic
modification removes an immunodominant allergen from soy-
bean. Plant Physiol. 2003,132,36-43.
(21) Vidic, M.; Hrustic, M.; Jockovic, D.; Miladinovic, J.; Dordevic,
V. 38th Seminar of Agronomists; Institute of Field and Vegetable
Crops: Novi Sad, Serbia and Montenegro, pp 129-139.
(22) Englyst, H. N.; Quigley, M. E.; Hudson, G. J.; Cummings, J. H.
Determination of dietary fibre as non-starch polysaccharides by
gas-liquid chromatography. Analyst 1992,117, 1707-1714.
(23) Copp, L. J.; Blekinsop, R. W.; Yada, R. Y.; Marangoni, A. G.
The relationship between respiration and chi color during long-
term storage of potato tubers. Am. J. Potato Res. 2000,77, 279-
287.
(24) Weiss, W.; Vogelmeier, C.; Gorg, A. Electrophoretic charac-
terization of wheat grain allergens from different cultivars
involved in bakers’ asthma. Electrophoresis 1993,14, 805-816.
(25) Blum, H.; Beier, H.; Gross, H. J. Improved silver staining of
plant proteins, RNA, and DNA in polyacrylamide gels. Elec-
trophoresis 1987,8,93-99.
(26) Munck, L.; Pram Nielsen, J.; Møller, B.; Jacobsen, S.; Sønder-
gaard, I.; Engelsen, S. B.; Nørgaard, L.; Bro, R. Exploring the
phenotypic expression of a regulatory proteome-altering gene
by spectroscopy and chemometrics. Anal. Chim. Acta 2001,446,
171-186.
(27) Refstie, S.; Sahlstro¨ m, S.; Bråthen, E.; Baeverfjord, G.; Krogedal,
P. Lactic acid fermentation eliminates indigestible carbohydrates
and antinutritional factors in soybean meal for Atlantic salmon
(Salmo salar). Aquaculture 2005,246, 331-345.
(28) Honig, D. H.; Rackis, J. J. Determination of the total pepsin-
pancreatin indigestible content (dietary fiber) of soybean prod-
ucts, wheat bran, and corn bran. J. Agric. Food Chem. 1979,
27, 1262-1266.
(29) Slominski, B. A. Hydrolysis of galactooligosaccharides by
commercial preparations of R-galactosidase and β-fructofuranosi-
dasesPotential for use as dietary additives. J. Sci. Food Agric.
1994,65, 323-330.
(30) Parsons, C. M.; Zhang, Y.; Araba, M. Nutritional evaluation of
soybean meals varying in oligosaccharide content. Poult. Sci.
2000,79, 1127-1131.
(31) Smiricky-Tjardes, M. R.; Flickinger, E. A.; Grieshop, C. M.;
Bauer, L. L.; Murphy, M. R.; Fahey, G. C. In Vitro fermentation
characteristics of selected oligosaccharides by swine fecal
microflora. J. Anim. Sci. 2003,81, 2505-2514.
(32) Smiricky, M. R.; Grieshop, C. M.; Albin, D. M.; Wubben, J.
E.; Gabert, V. M.; Fahey, G. C. The influence of soy oligosac-
charides on apparent and true ileal amino acid digestibilities and
fecal consistency in growing pigs. J. Anim. Sci. 2002,80, 2433-
2441.
(33) Wiggins, H. S. Nutritional value of sugars and related compounds
undigested in the small gut. Proc. Nutr. Soc. 1984,43,69-75.
Received for review June 17, 2005. Revised manuscript received
September 8, 2005. Accepted September 15, 2005.
JF051438R
Evaluation of NSP and Oligosaccharide Content of Soybeans
J. Agric. Food Chem.,
Vol. 53, No. 23, 2005 9121