Journal of Cereal Science 75 (2017) 92e99

Contents lists available at ScienceDirect

Journal of Cereal Science

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

Monitoring the staling of wheat bread using 2D MIR-NIR correlation

spectroscopy
Tine Ringsted a, *, Heinz Wilhelm Siesler b, Søren Balling Engelsen a
a
Chemometrics and Analytical Technology, Department of Food Science, University of Copenhagen, Rolighedsvej 26, 1958 Frederiksberg, Denmark
b
Department of Physical Chemistry, University of Duisburg-Essen, Schützenbahn 70, D 45117 Essen, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Staling of bread is a major source of food waste and efficient monitoring of it can help the food industry
Received 25 October 2016 in the development of anti-staling recipes. While the staling fingerprint in the mid-infrared region is
Received in revised form fairly well established this paper set out to find the most informative parts of the near-infrared spectra
15 February 2017
with respect to staling. For this purpose, two-dimensional correlation spectroscopy on near- and mid-
Accepted 12 March 2017
Available online 16 March 2017
infrared spectra of wheat bread crumb during aging was employed for the first time. The important
mid-infrared absorption band at 1047 cm
1 related to amylopectin retrogradation was found to correlate
positively with increased bread hardness and to co-vary with the near-infrared band at 910 nm in the
Keywords:
Near-infrared spectroscopy
short wavelength region (r2 ¼ 0.88 to hardness), the near-infrared band at 1688 nm in the 1. overtone
Mid-infrared spectroscopy region (r2 ¼ 0.97 to hardness) and to the near-infrared band in the long wavelength region at 2288 nm
2DCOS (r2 ¼ 0.97 to hardness). The spectral information from the first principal component on near-infrared and
Bread staling the first principal component on mid-infrared was found to be highly correlated by a r2 ¼ 0.98. It is
demonstrated that the major bread staling processes such as amylopectin retrogradation and water loss
can be followed with both near- and mid-infrared spectroscopy.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction including X-ray diffraction, differential scanning calorimetry (DSC),

mid-infrared spectroscopy (MIR) and proton nuclear magnetic
The non-microbial changes in bread during storage which lead resonance (NMR) relaxation (Gray and Bemiller, 2003). Of these
to consumer disapproval are called staling. Some of the staling methods, the most convenient is probably the vibrational spec-
processes are loss of flavor, increased hardness of the crumb and a troscopic method because of its fast measurement speed, little or
decreased freshness (Fadda et al., 2014). Stale bread contributes no sample preparation and it can measure several quality param-
significantly to the amount of food waste and many studies have eters simultaneously. MIR spectroscopy on bread has been used to
therefore investigated staling by varying bread constituents such as measure the change from a more amorphous starch structure in
maltogenic enzymes and production parameters. The physical- fresh bread to a more structured and crystalline starch in stale
chemical processes that contribute to the staling effect are com- bread by an intensity increase at 1047 cm
1 and an intensity
plex and still today not fully understood. However, some of the decrease at 1022 cm
1 (Sevenou et al., 2002; van Soest et al., 1995;
most important mechanisms are starch transformations, starch- Wilson et al., 1991). The changes in starch structure which can be
gluten interactions and moisture redistribution (Fadda et al., studied by MIR spectroscopy have been related to the double helix
2014; Gray and Bemiller, 2003). content and alignment in starch, the so-called short-range order
The industrial reference measurement to bread staling and the (Sevenou et al., 2002; Wilson et al., 1987). However, MIR spec-
most direct correlation to sensory evaluation is texture profile troscopy typically uses the attenuated total reflection (ATR) set-up
analysis (TPA) (Amigo et al., 2016; Bourne, 1978). Several other that provides a rather small sampling area and thus potentially
methods are available for measuring staling/starch retrogradation, non-representative sampling. ATR crystals with multiple re-
flections are available but the voids in the bread structure make it
difficult to obtain a good and reproducible contact with the
* Corresponding author. reflection element. The sampling area can be effectively increased if
E-mail addresses: [email protected] (T. Ringsted), [email protected] near-infrared (NIR) spectroscopy can be used instead (Engelsen,
(H.W. Siesler), [email protected] (S.B. Engelsen).

http://dx.doi.org/10.1016/j.jcs.2017.03.006
0733-5210/© 2017 Elsevier Ltd. All rights reserved.
 T. Ringsted et al. / Journal of Cereal Science 75 (2017) 92e99 93

2016). Some studies have used NIR spectroscopy on staling bread, to D2O. The bread crumb pieces were then exposed to air saturated
but most studies used raw absorbance spectra which contain both with D2O and MIR spectra was measured after 4, 24, 29, 47, 51 and
physical and chemical information (Abu-Ghoush et al., 2008; Cevoli 72 h. Two bread pieces were measured at each time point and the
et al., 2015; Wilson et al., 1991; Xie et al., 2004). If the physical spectra were averaged.
information in the spectra is removed, then NIR and MIR contain
similar chemical information based on molecular vibrations, 2.2. Texture profile analysis
although they may be influenced in slightly different ways by
intermolecular interactions (Engelsen, 2016). This work aims at TPA was performed with a TA.XT2 Texture Analyser (Stable
investigating if the NIR region can provide the same information Micro Systems Ltd., Surrey, UK). The applied probe was a 50 mm
about bread staling as is found in the MIR region. This will be aluminum cylinder probe (P/50, Stable Micro Systems Ltd., Surrey,
investigated by studying the assignment of MIR spectra of bread by UK). The TPA measurement was set to: pre-test speed 2 mm/s, test
using deuterium exchange experiments and by applying for the speed 2 mm/s, post-test speed 10 mm/s. Prior to the measurement,
first time two-dimensional (2D) correlation spectroscopy to the the probe was calibrated to the height of the bread sample and set
bread staling process with the aim of improving the assignments of to compress the sample 40% with a 5 kg load cell. The time interval
the vibrational modes that change during the staling process. The between measurement of the first and second peak was set to 5 s.
2D correlation spectroscopy on MIR and NIR spectra relates the The force was measured in Newton (N). Hardness was calculated as
fundamental vibrations in the MIR region with the overtones and the height of the first compression peak.
combination bands in the NIR region (Barton et al., 1992; Noda,
1993). This technique can be used to augment the interpretation 2.3. Spectroscopy measurements
of peaks which are well known in one dimension (e.g. MIR spectra),
but elusive in the other dimension (e.g. NIR spectra). 2.3.1. Mid-infrared spectroscopy
MIR spectra on bread were collected on a MB100 FT-IR spec-
2. Experimental trometer (Bomem, Quebec, Canada). Spectra were acquired in the
ATR mode with a triple-bounce diamond crystal (Durascope, SensIR
2.1. Preparation of breads Technologies, Danbury, CT, USA). The pressure applied to squeeze
the bread sample towards the diamond was 5 N/cm2. A total of 128
2.1.1. Breads for staling experiment scans were averaged for each sample and the spectral resolution
The flour applied in this project was kindly donated by Danisco was 4 cm
1. A clean ATR crystal was used as a background. Wheat
(Brabrand, Denmark). The flour was Reform Wheat Flour from starch, wheat gluten, granular/native amylose from barley (Carciofi
Denmark of medium quality. The water content in the flour was et al., 2012), granular/native amylopectin from maize and D-
~14%, protein content ~10% and ~0.5% ashes. Small amounts of a- glucose were measured with the same settings of the MIR spec-
amylase and ascorbic acid had been added to the flour in order to trometer as the bread. The deuterated bread was measured on an
improve the dough quality, especially the dough strength. The flour IFS 28 FT-IR spectrometer (Bruker Optik GmbH, Ettlingen, Ger-
was also added 1.5% salt and 1.5% sugar. A buffer (50 mM Sodium many) with a resolution of 4 cm
1 and averaging of 64 scans.
acetate, 2 mM CaCl2, pH 5.9) was added instead of water to Spectra were acquired using an ATR module with a single-bounce
assimilate the effect of yeast. diamond crystal (Durascope, SensIR Technologies, Danbury, CT,
The breads were prepared as follows. First all dry ingredients USA).
were mixed for 1 min at slow speed in a Diosna mixer. Then buffer
was added and the dough was mixed 2 min at low speed and 2.3.2. Near-infrared spectroscopy
4 min at high speed (temperature of the dough was 24e25  C). A 6500 NIR spectrometer (NIR systems, Inc., Silver Springs, MD,
Afterwards the dough was set to rest for 10 min at 30  C. Next the USA) was used to measure the bread in reflectance mode. The in-
dough was put to rest for 5 min at room temperature. Following strument contained a monochromator and a split detector system
this, the dough pieces were molded before they were put into tins where a Si detector was used between 400 and 1100 nm and a PbS
to proof for 60 min at 33  C at 85% relative humidity in a proofing detector was used between 1100 and 2498 nm. The measurement
cabinet. The loaves were baked for 30 min at 220  C with 12 s initial interval was every 2 nm. Angle of incident light was 180 and
steam in a Bago rotation oven with MIWE stone decks. The loaves reflectance was measured at a 45 angle. A rotating sample cup
were cooled for 3 h to make sure that the center temperature of the with a quartz window was used. A white ceramic plate was used as
breads reached room temperature. The finished loaves were packed a reference. The spectrum was an average of 32 scans. Wheat starch,
in vacuum bags and stored at ambient temperature. Right before wheat gluten, granular/native amylose from barley (Carciofi et al.,
the measurements, the bread was sliced into 10 mm slices using an 2012), granular/native amylopectin from maize and D-glucose
industrial slicing machine and a 50 mm cylinder was used to stick were measured on a NIRS™ DS2500 F (Foss, Hillerød, Denmark)
out a sample from the center of each bread slice. Two breads were equipped with a monochromator and Si and PbS detectors. The
measured at each time point. NIR and MIR spectroscopic mea- spectrum from DS2500 was an average of 256 scans measured from
surements on bread crumb were performed after 3 h, 9 h, 1, 2, 4, and 400 to 2500 nm with an interval of 0.5 nm.
7 days. Two bread slices from each end and one center slice was
measured. TPA was measured after 3 h, 1, 2, 4, and 7 days and ten 2.4. Data treatment
bread slices were measured from each bread and the average result
was used. The spectra were analyzed in Matlab (version R2013b) (The
Matworks, Inc., Natick, MA, USA) and PLS toolbox 7.5 (Eigenvector
2.1.2. Breads for deuterium experiment Research, Inc., Manson, WA, USA). Principal component analysis
Bread prepared from wheat flour, water, salt and yeast were (PCA) was used to visualize how samples where varying with
purchased from a local bakery few hours before measurements. respect to each other (Bro and Smilde, 2014).
From one bread, 14 crumb pieces of 2  2 cm were placed in a three-
neck angled round bottom flask. The first MIR spectrum (which was 2.4.1. Mid-infrared data treatment
referred to as 0 h) was taken from bread that had not been exposed The MIR spectra were calculated as log10(I0/Is) where I0 is the
94 T. Ringsted et al. / Journal of Cereal Science 75 (2017) 92e99

light intensity with no sample on the ATR and Is is the light in- 3. Results and discussion
tensity with a sample on the ATR. The data analysis was focused
only on the MIR region from 850 to 1100 cm
1 because of the 3.1. Assignment of the mid-infrared data
residing intense pyranosidic (ether) vibrations (van Soest et al.,
1995; Wilson et al., 1991). The effect of preprocessing with SNV From the raw MIR spectra in Fig. 1A, the small pyranosidic ether
(standard normal variate), MSC (multiplicative scatter correction), region from 850 to 1100 cm
1 was selected (in Fig. 1B) because of its
EMSC (extended multiplicative scatter correction) (Martens et al., established correlation to bread staling and amorphous/crystalline
2003) and SavitzkyeGolay 2nd derivative spectra (2nd degree structure of carbohydrates (van Soest et al., 1995; Wilson et al.,
polynomial, window size 7) was investigated (Rinnan et al., 2009). 1991). The region involves highly coupled vibrational modes that
cannot be assigned to specific functional groups. Table 1 shows an
overview of the suggested assignments of this region. The absor-
2.4.2. Near-infrared data treatment bance from amylopectin, amylose, starch, gluten and D-glucose was
The NIR spectra were calculated as log10(1/reflectance) where also measured in order to see which of these components could be
reflectance ¼ IR/IR0 and IR and IR0 are the reflected light from the responsible for the different absorbance peaks (Supplementary
sample and the reference, respectively. The focus in the NIR region Information Fig. S5). Fig. S5 shows that amylopectin and starch
was chosen as the short wavelength (SW) region from 900 to absorb in the same spectral areas which makes sense because
1050 nm, the first overtone (1OT) region from 1350 to 1800 nm and starch contains around 75% of amylopectin. It could also be
the long wavelength (LW) region from 2150 to 2370 nm. These observed that amylopectin, amylose and starch were the only
three NIR regions were chosen because of their information on OH species that absorb at 862 cm
1 which fits with the peak assign-
and CH vibrations. The effect of preprocessing with SNV, MSC, ment to the alpha-anomeric carbon C(1) in the glycosidic bond. It
EMSC (Martens et al., 2003) and SavitzkyeGolay 2nd derivative was surprising that amylose seemed to absorb at slightly higher
spectra (2nd degree polynomial, window size 11) was investigated wavenumbers than starch and amylopectin at 1022 and 1047 cm
1.
(Rinnan et al., 2009). The different preprocessing methods resulted Amylose and amylopectin are both polymers of a-D-glucose units
in dissimilar peak movements when applied to the full spectrum. with mainly a(1 / 4) glycosidic bonds but amylopectin also con-
On the other hand, similar peak movements were achieved when tains ~5% a(1 / 6) bonds (Pe
rez and Bertoft, 2010). The difference
the preprocessing methods were used on smaller spectral regions. between amylose and amylopectin at about 1047 and 1022 cm
1
It was therefore considered a more robust approach to proceed can be that the hydroxyl groups (COH) are involved in different
with the preprocessing of smaller regions because of the higher levels of intra- and inter-molecular hydrogen-bonding in amylose
agreement between the different preprocessing methods. The SW and amylopectin because of their single helical and double helical
region was therefore preprocessed from 850 to 1086 nm, the 1OT nature, respectively (Damager et al., 2010). Another interesting
region from 1270 to 1840 nm and the LW region from 2100 to observation is that amylose seemed to have a smaller absorbance at
2400 nm. 1047 cm
1 in comparison to (crystalline) amylopectin which sup-
ports the assignment to amylopectin retrogradation. From the raw
spectra in Fig. S5 it was observed that gluten had a relatively low
2.4.3. Two-dimensional correlation spectroscopy absorbance in this region compared to amylose, amylopectin and
The idea of 2D correlation spectroscopy was proposed by Noda starch. Furthermore, the protein content was around 8% in the
in 1986 and further developed into generalized 2D correlation bread and the effect from gluten is therefore not significant in the
spectroscopy (Noda, 1993). Barton and co-workers developed a region from 850 to 1100 cm
1. Deuteration of fresh bread during
similar technique (Barton et al., 1992). The technique can be used to approx. 3 days of storage showed that all the peaks at 1000, 1022,
augment the interpretation of peaks which are well known in one 1047 and 1078 cm
1 changed with deuteration (Fig. 2). This was
dimension, but elusive in the other dimension. It can also be used to expected since D2O will exchange the hydrogen atoms of water and
separate overlapping peaks by mapping samples that are influ- hydroxyl groups (COH), but not with the skeletal vibrations of CCH
enced by external perturbation (e.g. the samples vary in tempera- or OCH. The deuteration experiment therefore supports the strong
ture or in the chemical reaction time). The spectra can then be involvement of hydroxyl groups (COH) to the peaks at 1000, 1022,
visualized in plots showing how the wavelength variables co-vary 1047 and 1078 cm
1. In contrast, the peaks at 861 and 1011 cm
1
(in the synchronous plot) and change during the external pertur- did not show an intensity decrease in absorbance during the
bation (in the asynchronous plot). deuteration which supports their assignments to COC, CCH, CC,
In this study, the 2D correlation spectroscopy plots were OCH and CCO groups.
calculated as the covariance (also called the synchronous plot) The most efficient preprocessing method for separating storage
between NIR and MIR spectra. The covariance was calculated by the time or bread crumb hardness in a PCA on MIR spectra was found to
Matlab code (XNIR′*XMIR)/(n-1) where n is the number of samples, be 2DEV þ EMSC. Fig. 3A shows how the first principal component
XMIR and XNIR are the data matrixes with samples as rows and (PC1) displays an exponential decay with storage time and a linear
wavenumbers as columns for MIR and NIR, respectively (Czarnecki, decay with hardness (Fig. 3C). In the remainder of this study the
2011). XNIR′ is the transpose of XNIR. bread crumb hardness is used because of its linear relationship
The MIR spectra used for the 2D correlation spectroscopy were with the MIR spectra. Since the PCA input was 2nd derivative
preprocessed with EMSC from 850 to 1100 cm
1. The NIR spectra spectra then the interpretation of the loading plots in Fig. 3B and D
used for the 2D correlation spectroscopy were preprocessed with must be inverted when compared to the raw spectra. From the
EMSC from 850 to 1086 nm, 1270e1840 nm and 2100e2400 nm loading plot it can therefore be observed that raw spectra of the
(see Section 2.4.2 for an explanation of the small preprocessing new bread have higher values at 1084, 1065, 1024 and 995 cm
1
regions). The 2D correlation spectroscopy plots were very similar whereas the old “stale” bread have higher absorbance values at
when the MIR and NIR spectra were preprocessed with EMSC or 1074, 1049, 1007 and 862 cm
1. The peaks at 1084 and 1074 cm
1
with 2nd derivative spectra followed by EMSC (2DEV þ EMSC). It identified by the loading plot, have not received much attention in
was therefore decided to preprocess with only EMSC because the previous studies on bread or starch. However, one study found that
2D correlation spectroscopy plots using EMSC were simpler. Pre- the absorbance in this region did not show significant changes with
processed spectra were mean centered before modeling. varying degree of crystallinity in starch-water solutions (van Soest
 T. Ringsted et al. / Journal of Cereal Science 75 (2017) 92e99 95

Fig. 1. (A) Raw MIR spectra of aging bread crumb highlighting the area shown in Fig. 1B. (B) Spectra of selected MIR region preprocessed with EMSC. (C) Raw NIR spectra of aging
bread crumb highlighting the SW (900e1050 nm), the 1OT (1350e1800 nm) and the LW (2150e2370 nm) NIR region. The arrows indicate the spectral movement from 3 to 168 h.

Table 1 1049 cm
1 and oppositely more amorphous starch and fresh bread
MIR peak assignments. with a high absorbance at 1024 cm
1, which fits perfectly with
Wavenumbers Functional groups Sensitive to Reference previous studies (van Soest et al., 1995; Wilson et al., 1991). The
[cm
1] deuteration absorption band at around 1007 cm
1 has also been observed to
(Fig. 2) increase in previous studies on aging amylose and amylopectin gels
861 C(1)-O(5)-C(5) str., CCH No (Cael et al., 1975) (Goodfellow and Wilson, 1990) as well as on aging wheat-starch
1000 CO, CC, CCO, CCH, COH Yes (Cael et al., 1975) gels under drying conditions (Wilson and Belton, 1988). This
(Vasko et al., 1972)
matches with the assignment to CC, OCH, CCH and CCO since the
1011 CC, OCH, CCH, CCO No (Vasko et al., 1972)
1025e1026 COH def. Yes (Cael et al., 1975; migration of water from crumb to crust will result in a percentage
Vasko et al., 1972) increase of carbohydrates. However in another study, starch in
1045e1047 CO, CC, CCO, COH Yes (Cael et al., 1975) 0e25% water exhibited an increase in absorbance at around
(Vasko et al., 1972)
1000 cm
1 with both higher water content and higher amounts of
1076e1079 CO, CC, COH Yes (Cael et al., 1975)
(Vasko et al., 1972)
crystallinity (Capron et al., 2007). It is therefore a possibility that
the band at about 1007 cm
1 can be influenced by both crystallinity
and water migration. The peak at 862 cm
1 found in the loading
plot has also previously been observed to change with crystallinity
et al., 1995). The loading plot also reveal an association between
(van Soest et al., 1995).
more crystalline starch and stale bread with a high absorbance at

Fig. 2. MIR spectra of wheat bread crumb stored in an atmosphere saturated with D2O for 0e72 h. (A) Raw MIR spectra. (B) MIR spectra preprocessed with 2DEV þ EMSC. Black
arrows indicate the peak minimum.
96 T. Ringsted et al. / Journal of Cereal Science 75 (2017) 92e99

Fig. 3. Principal component 1 (PC1) from a principal component analysis (PCA) on MIR spectra of aging bread crumb preprocessed with 2DEV þ EMSC. (A) PC 1 scores versus time
and (B) the corresponding loading plot to Fig. 3A. (C) PC 1 scores versus bread crumb hardness and (D) the corresponding loading plot to Fig. 3C. Points are colored according to
hours after baking. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

In summary, this study supports the assignments of the peaks at be observed that the peak at 1047 cm
1 increase together with the
1047 and 1022 cm
1 to stale and fresh bread, respectively. The NIR wavelengths at 910, 1688 and 2288 nm. The SW-NIR band at
peaks at about 1000 and 1078 cm
1 both increased with storage 910 nm is the 3rd overtone of CH stretching presumably of protein
time and the absorbance at 1000 cm
1 might be related with water (gluten). The 1OT-NIR absorbance at 1688 nm is the aromatic CH
content and staling and the peak at 1078 cm
1 could be related to vibrations probably from gluten (Osborne et al., 1993; Williams,
the water content. The peak at 862 cm
1 increased with storage 2001). This is supported by the absorbance spectrum of gluten
time and was associated with stale bread. (Supplementary Information Figs. S6e8) which shows that the
absorbance at 910 and 1688 nm are superimposed by the CH vi-
brations in gluten. The LW-NIR band at 2288 nm is closest to the
3.2. Two-dimensional correlation spectroscopy on near- and mid- absorbance peak from amylose, but also starch, amylopectin and
infrared spectra gluten could contribute. The interpretation of the combination
band region is complex and the absorbance at 2288 nm could
The raw NIR spectra are shown in Fig. 1C and display a potentially be the combination band of the CH antisymmetric
decreasing absorbance from 3 to 168 h at all wavelengths. Three stretching þ CH deformation (Ringsted et al., 2016). The assignment
regions (indicated with gray areas) where selected from the NIR of the 910, 1688 and 2288 nm bands was confirmed by a deutera-
spectra in order to investigate how the SW, 1OT and LW region tion experiment on aging bread which showed almost no spectral
correlate to bread staling and the MIR spectra. NIR and MIR change or an increasing band intensity at the three wavelengths
spectra were used in 2D correlation spectroscopy and Fig. 4 (Supplementary Information Figs. S9e11). These findings support
shows the resulting covariance/synchronous plot. The absor- that OH or NH is not absorbing at these wavelengths.
bance bands which are moving in the same direction have a Fig. 4 also shows that the important MIR peak at 1022 cm
1
positive covariance (shown with red color) and absorbance bands decrease together with the NIR wavelengths at 974, 1412 and
moving in opposite directions have a negative covariance (shown 2258 nm. The absorbance at the SW-NIR band at 974 nm is the 2nd
with a blue color). It was also investigated if Pearson's correlation overtone of OH stretching due to water, but it may be influenced by
coefficient could be used instead of the covariance in the 2D OH stretching in starch, amylopectin and amylose (Supplementary
plots, but this resulted in less well defined peaks. Pearson's cor- Information Fig. S6). The 1OT band at 1412 nm is primarily related
relation coefficient sometimes results in square-looking correla- to the 1OT OH stretching from secondary hydroxyl groups which
tion peaks as for instance the correlation peak at 1502 nm in the primarily have absorbance from water, starch, amylopectin and
Supplementary Information Fig. S3C. Pearson's correlation coef- amylose (Supplementary Information Fig. S7). It is therefore hy-
ficient can also amplify the baseline as for example the area at pothesized that this decrease in absorbance (or rather shift towards
1600 nm in the Supplementary Information Fig. S3D and it was longer wavelengths) is caused mainly by a decrease in the average
therefore decided to use the covariance in the 2D correlation number (and/or strength) of polymer hydrogen bonding to water
plots. Since the covariance itself does not tell us how much the due to the lowering of moisture. The LW band at 2258 nm could be
NIR and MIR spectra correlate, then the Pearson's correlation caused by absorption from gluten, starch, amylopectin and amylose
coefficient (R2) was calculated between selected wavelengths and (Supplementary Information Fig. S8) and it has been assigned to OH
the bread hardness. The NIR and MIR preprocessing method stretching þ OH deformation (Osborne et al., 1993). It is also likely
which resulted in the highest R2 values was 2DEV þ EMSC and that water is contributing to the changes at 2258 nm since water
this preprocessing method was therefore used in the calculations starts to absorb from about 2230 nm and continues into the MIR
of R2 (Supplementary Information Figs. S1e4). spectrum (Law and Tkachuk, 1977).
The highest R2 to bread hardness are shown in Fig. 4 and it can
 T. Ringsted et al. / Journal of Cereal Science 75 (2017) 92e99 97

Fig. 4. MIR and NIR spectra of aging wheat bread crumb used in 2D correlation spectroscopy. The horizontal axis is the MIR region from 850 to 1100 cm
1 and the vertical axis is the
NIR regions from 900 to 1050 nm, 1350e1800 nm and 2150e2370 nm. The spectra were all preprocessed with EMSC. The colors represent the level of covariance where red is
positive and blue is negative. The Pearson's correlation coefficient (R2) was calculated between the spectra preprocessed with 2DEV þ EMSC (shown in Supplementary information
Fig. S2D, S3D and S4D) and bread crumb hardness. Arrows indicate spectral movement from 3 to 168 h. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)

3.3. Assignment of the long wavelength near-infrared during the staling process. The result is shown in Fig. 5B and has a
(2150e2370 nm) data Pearson's correlation coefficient of 0.98 which indicate that both
LW NIR and MIR spectroscopy basically measure the same staling
From Section 3.2 it was found that the highest R2 value between process and that they both can be used to describe changes in aging
hardness and a NIR wavelength was in the LW region at 2288 nm. bread crumb.
The correlations to the 1OT region from 1670 to 1800 nm were only
marginally lower, but the covariation around the LW band at
2288 nm is stronger than around the 1OT band at 1688 nm (Fig. 4). 4. Conclusions
It was then decided to examine the LW region from 2150 to
2370 nm in more details. A PCA built on the NIR region from 2150 to Bread staling is a major source of food waste and we are in
2370 nm is shown in Fig. 5A. The score of PC1 show a linear demand for more efficient methods to monitor staling. This work
decrease to bread crumb hardness as was also the case for the PCA represents the first application of 2D MIR-NIR correlation spec-
on the MIR spectra shown previously in Fig. 3C. troscopy to the bread staling process. The aim of the work was
Finally, the PC1 scores of the MIR and NIR spectra in Figs. 3C and primarily to investigate if NIR spectroscopy can be as informative
5A was correlated in order to see how much they were associated about the staling process as the MIR and to determine which region
of the NIR is the most appropriate for monitoring staling. The work
98 T. Ringsted et al. / Journal of Cereal Science 75 (2017) 92e99

Fig. 5. (A) Scatter plot of Principal component 1 (PC1) from a principal component analysis (PCA) on NIR spectra of aging wheat bread crumb preprocessed with 2DEV þ EMSC and
bread crumb hardness. (B) Scatter plot of NIR PC1 scores vs MIR PC1 scores. Points are colored according to hours after baking. (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)

starts by confirming the well-established correlation between the References

MIR absorbance at 1047 cm
1 and bread crumb hardness which is
interpreted as the retrogradation of amylopectin in aging bread. Abu-Ghoush, M., Herald, T.J., Dowell, F., Xie, F., Aramouni, F.M., Madl, R., 2008. Effect
of preservatives addition on the shelf-life extensions and quality of flat bread as
The 2D correlation spectroscopy shows that the MIR band at determined by near-infrared spectroscopy and texture analysis. Int. J. Food Sci.
1047 cm
1 moves together with three characteristic NIR bands Technol. 43, 357e364.
identified at 910 nm (SW-NIR), 1688 nm (1OT-NIR) and 2288 nm Amigo, J.M., Alvarez, A.D., Engelsen, M.M., Lundkvist, H., Engelsen, S.B., 2016. Staling
of white wheat bread crumb and effect of maltogenic alpha-amylases. Part 1:
(LW-NIR). Of these bands the strongest covariation to the 1047 spatial distribution and kinetic modeling of hardness and resilience. Food
band and highest correlation to the bread hardness was found at Chem. 208, 318e325.
the LW-NIR band at 2288 nm. Both the 1OT and the LW-NIR regions Barton II, F.E., Himmelsbach, D.S., Duckworth, J.H., Smith, M.J., 1992. Two-Dimen-
sional vibration spectroscopy: correlation of mid- and near-infrared regions.
exhibit R2 values of 0.97 to bread crumb hardness. The 1OT and LW Appl. Spectrosc. 46, 420e429.
region was therefore superior compared to the SW region in Bourne, M.C., 1978. Texture profile analysis. Food Technol. 32, 62e66.
monitoring the CH changes during bread retrogradation. In Bro, R., Smilde, A.K., 2014. Principal component analysis. Anal. Methods 6,
contrast, the other important MIR band at 1022 cm
1 displayed a 2812e2831.
Cael, J.J., Koenig, J.L., Blackwell, J., 1975. Infrared and raman spectroscopy of car-
high negative correlation to bread crumb hardness and was inter- bohydrates. Part VI: normal coordinate analysis of V-amylose. Biopolymers 14,
preted as COH def. possibly affected by different levels of hydrogen- 1885e1903.
Capron, I., Robert, P., Colonna, P., Brogly, M., Planchot, V., 2007. Starch in rubbery
bonding. The 2D correlation spectroscopy shows that the MIR
and glassy states by FTIR spectroscopy. Carbohydr. Polym. 68, 249e259.
absorbance at 1022 cm
1 co-varied with NIR absorbance bands
on, A., Holm, P.B.,
Carciofi, M., Blennow, A., Jensen, S.L., Shaik, S.S., Henriksen, A., Bule
from OH at 974 nm in the SW-NIR region, at 1412 nm in the 1OT- Hebelstrup, K.H., 2012. Concerted suppression of all starch branching enzyme
NIR region and at 2258 nm in the LW-NIR region. In this case, the genes in barley produces amylose-only starch granules. Bmc Plant Biol. 12, 223.
Cevoli, C., Gianotti, A., Troncoso, R., Fabbri, A., 2015. Quality evaluation by physical
SW-NIR region (900e1050 nm) was found to be marginally better tests of a traditional Italian flat bread Piadina during storage and shelf-life
than the 1OT and LW regions at describing the OH changes in starch improvement with sourdough and enzymes. Eur. Food Res. Technol. 240,
and water. In conclusion, this paper clearly demonstrates that the 1081e1089.
Czarnecki, M.A., 2011. Two-dimensional correlation analysis of hydrogen-bonded
NIR region equals the information content found in the MIR region systems: basic molecules. Appl. Spectrosc. Rev. 46, 67e103.
when it comes to staling of wheat bread. This is most clearly and Damager, I., Engelsen, S.B., Blennow, A., Møller, B.L., Motawia, M.S., 2010. First
convincingly shown by examining the high R2 value of 0.98 be- principles insight into the alpha-glucan structures of starch: their synthesis,
conformation, and hydration. Chem. Rev. 110, 2049e2080.
tween the first principal components from the staling sensitive Engelsen, S.B., 2016. Near infrared spectroscopy - a unique window of opportu-
regions in the MIR and NIR in Fig. 5B. nities. NIR news 27, 14e17.
Fadda, C., Sanguinetti, A.M., Del Caro, A., Collar, C., Piga, A., 2014. Bread staling:
updating the view. Compr. Rev. Food Sci. Food Saf. 13, 473e492.
Goodfellow, B.J., Wilson, R.H., 1990. A Fourier transform IR study of the gelation of
Acknowledgements amylose and amylopectin. Biopolymers 30, 1183e1189.
Gray, J.A., Bemiller, J.N., 2003. Bread staling: molecular basis and control. Compr.
Rev. Food Sci. Food Saf. 2, 1e21.
The authors wish to acknowledge the generous financial sup- Law, D.P., Tkachuk, R., 1977. Near-infrared diffuse reflectance spectra of wheat and
port from The Danish National Advanced Technology Foundation wheat components. Cereal Chem. 54, 256e265.
Martens, H., Nielsen, J.P., Engelsen, S.B., 2003. Light scattering and light absorbance
(now Innovation Fund Denmark) to the project entitled “Light & separated by extended multiplicative signal correction. Application to near-
Food”. Thanks also go to DuPont (5184-00012B) for supporting the infrared transmission analysis of powder mixtures. Anal. Chem. 75, 394e404.
research with breads and reference analysis and to Andreas Blen- Noda, I., 1993. Generalized two-dimensional correlation method applicable to
infrared, raman, and other types of spectroscopy. Appl. Spectrosc. 47,
now for providing the granular amylose and amylopectin samples.
1329e1336.
Osborne, B.G., Fearn, T., Hindle, P.H., 1993. Practical NIR Spectroscopy, with Appli-
cations in Food and Beverage Analysis, second ed. Longman Scientific and
Technical, Harlow, UK.
Appendix A. Supplementary data
rez, S., Bertoft, E., 2010. The molecular structures of starch components and their
Pe
contribution to the architecture of starch granules: a comprehensive review.
Starch - Sta €rke 62, 389e420.
Supplementary data related to this article can be found at http:// Rinnan, Å., van den Berg, F., Engelsen, S.B., 2009. Review of the most common pre-
dx.doi.org/10.1016/j.jcs.2017.03.006.
 T. Ringsted et al. / Journal of Cereal Science 75 (2017) 92e99 99

processing techniques for near-infrared spectra. Trac-Trends Anal. Chem. 28, Williams, P., 2001. Implementation of near-infrared technology. In: Williams, P.,
1201e1222. Norris, K. (Eds.), Near-infrared Technology, in the Agricultural and Food In-
Ringsted, T., Dupont, S., Ramsay, J., Jespersen, B.M., Sørensen, K.M., Keiding, S.R., dustries, second ed. American Association of Cereal Chemists, St. Paul, MN, USA,
Engelsen, S.B., 2016. Near-infrared spectroscopy using a supercontinuum laser: p. 168.
application to long wavelength transmission spectra of barley endosperm and Wilson, R.H., Belton, P.S., 1988. A Fourier-transform infrared study of wheat starch
oil. Appl. Spectrosc. 70, 1176e1185. gels. Carbohydr. Res. 180, 339e344.
Sevenou, O., Hill, S.E., Farhat, I.A., Mitchell, J.R., 2002. Organisation of the external Wilson, R.H., Kalichevsky, M.T., Ring, S.G., Belton, P.S., 1987. A Fourier-transform
region of the starch granule as determined by infrared spectroscopy. Int. J. Biol. infrared study of the gelation and retrogradation of waxy-maize starch. Car-
Macromol. 31, 79e85. bohydr. Res. 166, 162e165.
van Soest, J.J.G., Tournois, H., de Wit, D., Vliegenthart, J.F.G., 1995. Short-range Wilson, R.H., Goodfellow, B.J., Belton, P.S., Osborne, B.G., Oliver, G., Russell, P.L., 1991.
structure in (partially) crystalline potato starch determined with attenuated Comparison of Fourier transform mid infrared spectroscopy and near infrared
total reflectance Fourier-transform IR spectroscopy. Carbohydr. Res. 279, reflectance spectroscopy with differential scanning calorimetry for the study of
201e214. the staling of bread. J. Sci. Food Agric. 54, 471e483.
Vasko, P.D., Blackwell, J., Koenig, J.L., 1972. Infrared and raman spectroscopy of Xie, F., Dowell, F.E., Sun, X.Z.S., 2004. Using visible and near-infrared reflectance
carbohydrates. Part II: normal coordinate analysis of alpha-D-glucose. Carbo- spectroscopy and differential scanning calorimetry to study starch, protein, and
hydr. Res. 23, 407e416. temperature effects on bread staling. Cereal Chem. 81, 249e254.