Food Measure (2017) 11:1909–1918

DOI 10.1007/s11694-017-9573-7

ORIGINAL PAPER

Determination of total phenolic content and antioxidant capacity

of blueberries using Fourier transformed infrared (FT-IR)
spectroscopy and Raman spectroscopy
Xiaozhen Zheng1 · Yaxi Hu1 · Elfi Anggreani1 · Xiaonan Lu1 

Received: 9 January 2017 / Accepted: 7 June 2017 / Published online: 9 June 2017
© Springer Science+Business Media, LLC 2017

Abstract  This study investigated the feasibility of using Introduction

Fourier-transform infrared (FT-IR) and Raman spectros-
copies to quantify total phenolic content and antioxidant The benefits of antioxidants to humans are mainly asso-
capacity of blueberries. Blueberry extracts were applied to ciated with the suppressing effects on free radicals [5].
determine total phenolic content (Folin–Ciocalteu assay) Excessive production of reactive oxygen species (ROS)
and antioxidant capacity [oxygen radical absorbance capac- may result in oxidative stress to cellular lipids, proteins,
ity (ORAC) and 2,2-diphenyl-1-picrylhydrazyl (DPPH)], and nucleic acids, which play a significant role in the devel-
while pressed juices were used for spectroscopic analysis. opment of cancer, cardiovascular disorder, diabetes, central
Six partial least squares regression models with cross-val- nervous system diseases, and chronic obstructive pulmo-
idation were developed using FT-IR and Raman spectra nary diseases [5, 20]. Blueberries are well known as “super
of blueberries from 11 locations and their corresponding fruit” with a rich resource of minerals, phenolics, vitamins,
chemical testing values, followed by prediction tests using and dietary fibers that can potentially quench ROS [28] and
samples from another five locations. FT-IR prediction show high level of antioxidant capacity [5, 6]. The antioxi-
models show relatively good prediction power for ORAC, dant properties of blueberries are due to the high content
DPPH, and Folin–Ciocalteu values ­(R2-prediction = 0.6359, of flavonoids, including flavones, isoflavones, flavanones,
0.6580, and 0.8092; RMSE-prediction  = 6.19 and anthocyanins, and catechins [6, 23, 28]. Flavonoids per-
0.71  µmol trolox equivalents/g fresh weight, and 0.14  µg form their antioxidant function through scavenging free
GAE/g fresh weight), but Raman prediction models did not radicals, binding to metal ions, and inhibiting free radical-
yield a satisfactory result. FT-IR spectroscopy may be used generating enzymes [8]. Genotype, ripening stage, cultiva-
to rapidly determine blueberry antioxidants. tion technique, climate condition, and storage condition can
affect antioxidants in blueberry [27].
Keywords  Blueberry · Phenolics · Antioxidant · Infrared Different chemical assays have been applied for the
spectroscopy · Raman spectroscopy examination of phenolic content and antioxidant capacity,
such as Folin–Ciocalteu (FC), 2,2-diphenyl-1-picrylhydrazl
Xiaozhen Zheng and Yaxi Hu have equally contributed as co-first radicals (DPPH), and oxygen radical absorbance capacity
authors. (ORAC) assays. These assays can be categorized into two
major types: hydrogen atom transfer (HAT) and electron
Electronic supplementary material  The online version of this transfer (ET). HAT involves a competitive reaction scheme
article (doi:10.1007/s11694-017-9573-7) contains supplementary
material, which is available to authorized users.
where the tested antioxidants will compete with the sub-
strate for the peroxyl radicals [10] while ET detects the
* Xiaonan Lu capability of antioxidants to transfer one electron to reduce
[email protected] free radicals, metals, and/or carbonyls [14]. However, all
1
Food, Nutrition and Health Program, Faculty of Land
these chemical assays are time consuming, expensive, labor
and Food Systems, The University of British Columbia, intensive, and use a large amount of organic solvents.
Vancouver V6T 1Z4, BC, Canada

13
Vol.:(0123456789)

1910 X. Zheng et al.

Coupled with multivariate statistical analysis, both until further analysis. The extracts were centrifuged at
infrared and Raman spectroscopies have been applied for 13,000 × rpm (rotor: fiberlite f15-8 × 50  cy, Sorvall Leg-
the determination of phenolic and antioxidant profiles of end X1R, Thermo Scientific) for 20 min. Each supernatant
numerous food commodities, such as white wine [7], green was obtained through filtration with syringe and Millex
tea [29], onion [14, 19], and fruits [11]. Both spectroscopic ­Express® PES filter membrane (0.22  µm) and stored in
techniques measure the vibrational modes of molecules the dark at −80 °C for further analysis. The end product of
and can fingerprint the chemical composition of a complex extraction was referred to as blueberry extract.
matrix, such as foods. Specifically, infrared spectroscopy The remaining ground blueberry samples were homog-
records the absorption signals while Raman spectroscopy enized and pressed through sterile gauze to receive crude
collects the scattering signals [13]. juices. The crude juices were filtered with syringe and
To date, no study has been conducted to perform Raman Millex ­Express® PES filter membrane (0.22 µm) and stored
spectroscopy for the quantification of phenolic content in the dark at −80 °C for further analysis. The end product
and antioxidant capacity in berries or to comprehensively was referred to as blueberry juice.
evaluate the performance of Raman and infrared spectro-
scopic methods. In this study, we determined the total phe-
nolic content and antioxidant capacity of blueberries from Determination of total phenolic content using FC assay
different countries through the use of both infrared and
Raman spectroscopic models developed based on partial- Total phenolic content of each blueberry extract was deter-
least squares (PLS) regression algorithm. In addition, the mined using the procedures of the FC assay reported in a
functional groups that contribute to antioxidants in blue- previous work with minor modification [9]. Briefly, 50 μL
berry were studied using spectroscopic based PLSR load- blueberry extract, 3  mL deionized water, and 500  μL FC
ing plots. reagent were mixed for 8  min at 20 °C, followed by the
addition of 1.5  mL saturated sodium carbonate. The mix-
ture was incubated in the dark at 20 °C for 2 h before meas-
Materials and methods uring the absorbance at 765 nm using a Shimadzu UV–Vis-
ible spectrophotometer. Blank sample was deionized water.
Chemical and reagents Total phenolic content values were determined using a cali-
bration curve prepared with gallic acid standards (0, 0.05,
FC reagent, DPPH, 6-hydroxy-2,5, 7,8-tetramethyl-2-car- 0.1, 0.15, 0.30, 0.45, 0.60, 0.75, and 1.00  mg/L). Results
boxylic acid (trolox), gallic acid, 2,2′-azobis (2-methyl- were expressed as mg of gallic acid equivalents/g fresh
propionamidine) dihydrochloride (AAPH), and fluorescein weight (mg GAE/g FW) blueberry. Each extract was deter-
sodium salt were obtained from Sigma-Aldrich (St. Louis, mined at least in triplicate.
MO, USA). Sodium carbonate ­ (Na2CO3), sodium phos-
phate monobasic anhydrous ­(NaH2PO4), potassium phos-
phate dibasic (­K2HPO4), and methanol were purchased Determination of total antioxidant capacity using
from Fisher Scientific (Ottawa, Ontario, Canada). All rea- DPPH Assay
gents and solvents used were analytical or HPLC grade.
Deionized water (18.2  MΩ/cm) was received from a Mil- The antioxidant capacity of blueberry extracts was deter-
lipore system at The University of British Columbia. mined using a DPPH assay as described by Hu and col-
leagues in a previous study [9] with minor revisions. DPPH
Blueberry sample preparation stock solution was prepared by dissolving 15  mg DPPH
in 30  mL 70% methanol and stored at −20 °C until use.
A total of 16 boxes of blueberry samples were received DPPH working solution was obtained by diluting stock
during the period from 2014 to 2015. These blueberries solution with 70% methanol to obtain an absorbance of
were produced from different farms originated at various 1.1 ± 0.1 at 515  nm using a spectrophotometer [27]. An
locations in Argentina, Chile, Canada, and USA. All sam- aliquot (0.1  mL) of each blueberry extract was added to
ples were stored in the dark at −80 °C until analysis. 2  mL DPPH working solution. Then, the absorbance was
Blueberry samples were thawed and ground with mor- measured at 515 nm after incubation for 30 min in the dark
tar and pestle. A total of 2.0  g (±0.02  g) of each sample at 20 °C. Trolox dilutions in 70% methanol in a range of
was extracted with 70% aqueous methanol (15  mL × 3) 0–800 μmol/L were used as the calibration standards. Total
in an ultrasonic water bath (Fisher Scientific) for 15  min antioxidant capacity of blueberry was expressed as μmol
each time for three times at 20 °C. The remaining ground trolox equivalents/g fresh weight (μmol trolox/g FW) blue-
blueberry samples were stored in the dark at −80  °C berry. Each extract was determined at least in triplicate.

13
Determination of total phenolic content and antioxidant capacity of blueberries using Fourier… 1911

Determination of total antioxidant capacity using Multivariate statistical analysis

ORAC Assay
FT-IR spectral fingerprinting region (2000–900  cm−1)
Phosphate buffer (75  mM, pH 7.4) prepared by ­NaH2PO4 was selected for further analysis because the target func-
and ­K2HPO4 was used to dilute samples and reagents. tional groups in blueberry juice can generate spectral
Samples and trolox with a series of dilutions (100 μL) was bands at this wavenumber region while noises domi-
mixed with 60  μL fluorescein (200  nmol/L) and 40  μL nate other wavenumber regions. The whole wavenumber
AAPH (60 mmol/L) in a 96-well plate. Fluorescence with region of Raman spectra (400–1800  cm−1) was selected
the excitation of 485 nm and the emission of 527 nm was for further analysis. Automatic baseline correction was
read every minute for 60 min by using a fluorescent micro- applied to both FT-IR and Raman spectra using OMNIC
plate reader (Tecan Austria GmbH, Tecan infinite M200 v7.0 (Thermo-Nicolet, Madison, WI, USA).
Pro). A linear relationship between the area under the curve A total of 16 blueberry samples from different geo-
for blueberry sample or trolox and its corresponding con- graphical regions were divided into both calibration set
centration was determined. The value of ORAC is equiva- and prediction set. Eleven samples were assigned as
lent to the slope of sample divided by the slope of trolox the calibration set and the remaining five samples were
and the result was expressed as μmol of trolox equivalents/g assigned as the prediction set for evaluating the perfor-
fresh weight (μmol trolox/g FW) blueberry. Each blueberry mance of the calibration models. Specifically, samples
extract was determined at least in triplicate. containing the highest and lowest values of the chemical
profiles as determined by FC, DPPH, and ORAC assays
were allocated into the calibration set and the remaining
FT‑IR spectral collection samples were randomly allocated into both calibration set
and prediction set.
FT-IR spectra of each blueberry juice sample were col- The calibration models were constructed to correlate
lected at 20 °C using a Perkin Elmer FT-IR spectrometer FT-IR or Raman spectra and the values of total phenolic
equipped with an attenuated total reflectance sensing sys- content or antioxidant capacity determined by traditional
tem (PerkinElmer, Waltham, MA, USA). Blueberry juice chemical-based assays using PLS regression algorithm in
samples were scanned over the wavenumber region of Matlab 2014b (MathWorks Inc., Natick, MA, USA). A
4000–550  cm−1 at a resolution of 4  cm−1 against a back- total of six PLS regression models were built in the end
ground of air. Each spectrum was an average of 16 scans, with the combination of two instrumental datasets (i.e.
which was collected using SPECTRA software (Perki- FT-IR and Raman spectra) and three chemical datasets
nElmer, Waltham, MA, USA). An aliquot of 10 µL of blue- (i.e. FC, DPPH, and ORAC assays). Internal validation
berry juice was applied onto the attenuated total reflectance within the calibration model was conducted using tenfold
sensing system of the FT-IR spectrometer and a total of 10 cross-validation: one-tenth of the data in the calibration
spectra of were collected for each sample. set was sequentially removed and treated as external to
validate the model built by the remaining data [3]. Cross
validation aids in selecting the optimal number of the
Raman spectral collection
latent variables (m) with two criteria, namely minimizing
the root mean square error of cross-validation (RMSECV)
A Renishaw confocal Raman spectroscopic system
and avoiding over-fitting. RMSECV is calculated using
equipped with a 785  nm near-infrared laser (maximum � �2
30 mW incident laser power) and a charge coupled device the equation: RMSECV = N v=1 i=1 yiv − ŷ iv
1 ∑v ∑n �

(CCD) array detector was used for collecting Raman spec- [25], where N is the number of samples in the training
tra. The settings of the Raman spectrometer include the set; v is the number of segments; n is the number of sam-
entrance aperture at 50  µm, focal length at 300  mm, and ples in a given segment; ­yiv is the reference value for
a 1200 line m­ m−1 grating. The CCD array detector (578- sample i in segment v, and ŷiv is the predicted value for
pixel by 385-pixel) has a pixel size of 22 µm. Raman spec- sample i in segment v. After establishing the linear
tra were collected from 400 to 1800 cm−1 at 20 °C. A total regression models between the spectral features and the
of ten spectra were collected from different locations from chemical results with the optimized number of latent var-
each blueberry juice sample. iables, the prediction power of these calibration models
were evaluated using the five samples in the prediction
set. Performance of each PLS regression model was
assessed based upon the calculated correlation of deter-
mination ­(R2), root mean squares error of calibration

13

1912 X. Zheng et al.

(RMSEC), root mean squares error of cross-validation FT‑IR spectroscopy and PLS regression models
(RMSECV), and root mean squares error of prediction for the determination of total phenolic content
(RMSEP). and antioxidant capacity of blueberries

A total of ten FT-IR spectra were collected per blueberry

sample; therefore, 160 spectra were collected and they
Results and discussion are shown in Fig.  1a. Band assignment of FT-IR spectra
of blueberries are summarized in Table  S2 [17, 22, 23].
Phenolic and antioxidant profiles of blueberries There was only one band (at 3360 cm−1) shown up between
the wavenumber region of 4000–2000  cm−1, which was
The total phenolic content and antioxidant capacity of mainly due to the –OH stretching of water in sample.
blueberries from different geographical locations are Moreover, there was noise in the wavenumber region of
summarized in Table  S1. These results were compara- 900–500  cm−1. As a consequence, only the fingerprinting
ble to a few previous studies. Specifically, the ORAC region (i.e. 2000–900 cm−1) was selected for baseline cor-
values ranging between 13.9 and 118.7  µmol trolox rection and further spectral analysis (Fig. 1b).
equivalents/g FW for blueberry samples were reported The baseline corrected fingerprinting spectra were
[11, 20, 23, 28]. In general, blueberries harvested from applied for the construction of PLS regression models with
different farms originated from different countries (i.e., the corresponding ORAC, DPPH, and FC values, respec-
Argentina, Chile, Canada and USA) have different pro- tively. Eleven samples were selected to be the calibration
files of phenolic content and antioxidant capacity. This set (Fig.  2) while the remaining five samples as the test
variation was due to the differences in genotypes, ripen- set. The statistical parameters of PLS regression models
ing age, cultivation methods, and climate conditions [23]. for the determination of total phenolic content and antioxi-
In addition, the application of different chemical assays dant capacity of blueberries are summarized in Table 1. A
can result in variations in the values of phenolic content strong PLS regression model has a relatively high regres-
and antioxidant capacity, and no correlation was discov- sion coefficient ­(R2 close to 1) with preferably a small
ered between the results obtained from DPPA, ORAC number of latent variables and root mean squares error
and FC assays (data not shown). Such variation might [14]. The calibrated models for all three chemical assays
be due to the different principles involved in the three (i.e. ORAC, DPPH, and FC) demonstrated high correla-
aforementioned chemical assays. First, the ORAC assay tion between FT-IR spectra and the chemical assays. How-
is a HAT-based reaction providing information on radi- ever, the performance of the models for cross-validation
cal chain-breaking capacity, while DPPH and FC assays and prediction was less satisfactory, with R ­ 2 of cross-val-
are based upon the ET reactions reflecting the reducing idations to be 0.7397, 0.7459, and 0.8373, and predictions
capacity of the samples [21]. It has been reported that to be 0.6359, 0.6580, and 0.8092 for ORAC, DPPH, and
many ORAC-active antioxidants are not sensitive to the FC assays, respectively. PLS regression calibration models
stable DPPH free radicals (e.g. chlorogenic acid), thus resulted in the best parameters for the FT-IR spectra against
DPPH assay could result in a lower antioxidant capac- FC assay and this could be attributed to the nature of the
ity [10, 12]. Besides, being conducted under basic con- FC reaction. ORAC and DPPH assays have relatively short
ditions, simple phenolics could react with FC reagent reaction/monitoring time (i.e. 60 and 30 min, respectively),
but lack the capacity to scavenge the free radicals (e.g. while FC assay requires 2  h to complete. The shorter the
DPPH). Therefore, even though involving the same reac- reaction/monitoring time, the more random error could be
tion mechanism, total phenolic results (i.e. FC assay) induced during operation, resulting in the reduced accu-
might not have high correlation with antioxidant capacity racy. Moreover, conducted under basic conditions, activ-
determined by DPPH assay. Second, ORAC assay meas- ity of antioxidants toward FC reagent increased [10], thus
ures fluorescence to determine both inhibition degree the relative random error and systematic error could be
and time, while DPPH and FC assays simply measures minimized and the results could be more representative to
the inhibition effect at a certain time, which might not be reflect the antioxidant composition of the overall sample.
representative [4, 12, 23]. Lastly, DPPH assay is a colori- The loading plots of PLS regression models (Figure
metric assay that is dependent upon the color change at S1) illustrated the contribution of each wavenumber in
515  nm, which is the maximum UV–Vis absorption for FT-IR spectra to the correlation relationship. The band at
DPPH radicals [10]. However, anthocyanin compounds in ~1650  cm−1 (assigned to aromatic C=C vibration in phe-
blueberries also exhibit absorption at 500–550 nm, which nolics) in the ORAC model contributed to the highest
could disturb the accuracy of DPPH assay [26]. positive correlation. The band at 1020  cm−1 (assigned to
O–H group in carbohydrate) in both DPPH and FC models

13
Determination of total phenolic content and antioxidant capacity of blueberries using Fourier… 1913

Fig.  1  a 160 raw FT-IR spectra

of blueberry (500–4000 cm−1);
b 160 baseline corrected
FT-IR spectra of blueberry at
fingerprinting region (900–
2000 cm−1)

contributed to the highest negative correlation, while the shown in Fig. 3a and the baseline corrected Raman spectra
band at 1080 cm−1 (assigned to aromatic CH bending and are shown in Fig. 3b. Band assignment of Raman spectra of
rocking and C–OH bending) contributed to the highest pos- blueberries are summarized in Table S3 [2, 15, 16, 22, 24].
itive correlation [17]. For the construction of PLS regression models, the
blueberry samples used for calibration set and prediction
Raman spectroscopy and PLS regression models set were the same as the ones used in FT-IR spectroscopic
for the determination of total phenolic content models (Fig. 4). The statistical parameters of Raman spec-
and antioxidant capacity of blueberries troscopic-based PLS regression models for the determina-
tion of total phenolic content and antioxidant capacity of
Similar to FT-IR spectroscopic analysis, ten Raman spectra blueberries are summarized in Table  2. The calibration
were collected per blueberry sample. A total of 160 Raman models for both DPPH and FC assays yield a relatively
spectra in the wavenumber region of 1800–400  cm−1 are high regression coefficient of 0.93 and 0.96, respectively,

13

1914 X. Zheng et al.

Fig.  2  a FT-IR calibration

model for ORAC assay; b
FT-IR calibration model for
DPPH assay; c FT-IR calibra-
tion model for FC assay

13
Determination of total phenolic content and antioxidant capacity of blueberries using Fourier… 1915

Table 1  PLS regression models for the determination and prediction of total phenolic content and antioxidant capacity of blueberries using
FT-IR spectroscopy
N n LV RMSEC R2 RMSECV R2-CV RMSEP R2-P

ORAC 110 50 7 2.8125 0.9195 4.6676 0.7397 6.1854 0.6359

DPPH 110 50 9 0.4355 0.9576 0.9689 0.7459 0.7069 0.6580
FC 110 50 7 0.1933 0.9307 0.2857 0.8373 0.1434 0.8092

N number of spectra for calibration, n number of spectra for prediction, LV latent variables, RMSEC root mean squares error of calibration,
RMSECV root mean squares error of cross-validation (tenfold), RMSEP root mean squares error of prediction

Fig.  3  a 160 raw Raman

spectra of blueberry (400–
1800 cm−1); b 160 baseline
corrected Raman spectra of
blueberry (400–1800 cm−1)

13

1916 X. Zheng et al.

Fig.  4  a Raman calibration

model for ORAC assay; b
Raman calibration model for
DPPH assay; c Raman calibra-
tion model for FC assay

13
Determination of total phenolic content and antioxidant capacity of blueberries using Fourier… 1917

Table 2  PLS regression models for the determination and prediction of total phenolic content and antioxidant capacity of blueberries using
Raman spectroscopy

N n LV RMSEC R2 RMSECV R2-CV RMSEP R2-P

ORAC 110 50 5 4.1370 0.7998 5.9723 0.3507 4.3026 0.2874

DPPH 110 50 5 0.5656 0.9260 0.8059 0.6474 0.9192 0.1481
FC 110 50 6 0.1455 0.9620 0.2749 0.6627 0.3501 0.1066

N number of spectra for calibration, n number of spectra for prediction, LV latent variables, RMSEC root mean squares error of calibration,
RMSECV root mean squares error of cross-validation (tenfold), RMSEP root mean squares error of prediction

with satisfactory number of latent variables. However, to predict the anthocyanins in mature blueberry compared
none of the prediction models resulted in a good prediction to the prediction capability towards unripen fruits. How-
power on the basis of ­R2 for prediction and the RMSEP. ever, in this current study, all the blueberry samples were
Furthermore, the loading plot (Figure S2) did not indicate in their mature states and the failure of using Raman spec-
any specific wavenumber region that contributes to a posi- troscopy to quantify antioxidant capacity and total phenolic
tive or negative correlation to the results of the traditional content is in accordance with the results found by Arrobas
chemical assays. Taken together, Raman spectroscopic and the colleagues.
based method was not feasible to determine antioxidant
capacity and total phenolic content in blueberry.
Only one study has been published using Raman spec- Conclusion
troscopy to determine the concentration of anthocyanins in
blueberry at different stages of ripeness and concluded that Both FT-IR and Raman spectroscopies were applied to
Raman spectroscopic method was feasible to quantify the replace the traditional chemical assays (i.e. DPPH, ORAC
anthocyanins in blueberry [1]. However, the authors only and FC assay) for the determination of antioxidant capac-
assessed the calibration and cross-validation models for ity and total phenolic content of blueberry. Only FT-IR
PLS regression and no external unknown sample was used spectroscopy was validated to be feasible to provide accept-
to validate the PLS regression models. Besides, the cross- able accuracy for the prediction of the antioxidant values
validation method used in the aforementioned study was determined by the three traditional assays and the overall
“leave-one-out”, which has higher possibility to overfit the analysis can be finished in a few minutes instead of hours
model, resulting in less robustness compared to “k-fold” or days required by traditional tests. Raman spectroscopy
cross-validation method [18]. Therefore, no evidence has failed to predict the antioxidant capacity and total phenolic
been found to demonstrate the success of applying Raman content of mature blueberries, but further research could be
spectroscopy to quantify the antioxidants in blueberry. conducted to apply Raman spectroscopy for assessing the
The failure of using Raman spectroscopy to study the ripeness of blueberry.
antioxidant capacity and total phenolic compound could
be attributed to the interference of high florescent signals Acknowledgements  This study was supported by Discovery,
developed from the biological samples (i.e. blueberry), Engage, and Collaborative Research and Development Grants
awarded to X. L. by Natural Sciences and Engineering Research
which is evidenced in Fig. 3. The 785-nm laser has higher Council of Canada. X. L. also thanks the financial support by Mitacs
tendency to induce the florescent signals compared to the Accelerate Program in Canada.
1024-nm laser used by Arrobas et al. [1]. Although baseline
correction was applied to remove the florescent signals in
each spectrum, the weak Raman signals could be affected, References
resulting in reduced sensitivity of Raman spectroscopy
towards compounds at low concentrations. Besides, the 1. B.G. Arrobas, L. Ciaccheri, A.A. Mencaglia, F.J. Rodriguez-
laser power of the Raman spectroscopic system used in our Pulido, C. Stinco, M.L. González-Miret, F.J. Heredia, A.G. Mig-
nani, Raman spectroscopy for analyzing anthocyanins of lyo-
current study was 30 mW, which might be enough to have
philized blueberries, in SENSORS, 2015 IEEE, 1–4 Nov 2015,
high sensitivity towards compounds at low concentrations. pp. 1–4
Moreover, as concluded by Arrobas et al. [1], large amount 2. M. Baranska, H. Schulz, E. Joubert, M. Manley, In situ flavonoid
of emerging compounds other than anthocyanins could be analysis by FT-Raman spectroscopy: identification, distribution,
and quantification of aspalathin in green rooibos (Aspalathus lin-
presented during blueberry maturation, masking the Raman
earis). Anal. Chem. 78(22), 7716–7721 (2006)
signals of anthocyanins at relatively low concentrations, 3. K. Baumann, M. von Korff, H. Albert, A systematic evaluation
thus Raman spectroscopy demonstrated worse performance of the benefits and hazards of variable selection in latent variable

13

1918 X. Zheng et al.

regression. Part II. Practical applications. J. Chemometrics 16(7), 16. Z. Movasaghi, S. Rehman, I.U. Rehman, Raman spectroscopy of
351–360 (2002) biological tissues. Appl. Spectrosc. Rev. 42(5), 493–541 (2007)
4. A. Brambilla, R. Lo Scalzo, G. Bertolo, D. Torreggiani, Steam- 17. Z. Movasaghi, S. Rehman, D.I. ur Rehman, Fourier transform
blanched highbush blueberry (Vaccinium corymbosum L.) juice: infrared (FTIR) spectroscopy of biological tissues. Appl. Spec-
phenolic profile and antioxidant capacity in relation to cultivar trosc. Rev. 43(2), 134–179 (2008)
selection. J. Agric. Food. Chem. 56(8), 2643–2648 (2008) 18. AY Ng, Preventing “overfitting” of cross-validation data, in

5. K. Dastmalchi, G. Flores, V. Petrova, P. Pedraza-Peñalosa, E.J. ICML, 1997, pp. 245–253
Kennelly, Edible neotropical blueberries: antioxidant and com- 19. Y. Numata, H. Tanaka, Quantitative analysis of quercetin using
positional fingerprint analysis. J. Agric. Food. Chem. 59(7), Raman spectroscopy. Food. Chem. 126(2), 751–755 (2011)
3020–3026 (2011) 20. R.L. Prior, G. Cao, A. Martin, E. Sofic, J. McEwen, C. O’Brien,
6. G. Flores, K. Dastmalchi, A.J. Dabo, K. Whalen, P. Pedraza- N. Lischner, M. Ehlenfeldt, W. Kalt, G. Krewer, Antioxidant
Peñalosa, R.F. Foronjy, J.M. D’Armiento, E.J. Kennelly, Anti- capacity as influenced by total phenolic and anthocyanin con-
oxidants of therapeutic relevance in COPD from the neotropical tent, maturity, and variety of Vaccinium species. J. Agric. Food.
blueberry Anthopterus wardii. Food. Chem. 131(1), 119–125 Chem. 46(7), 2686–2693 (1998)
(2012) 21. R.L. Prior, X. Wu, K. Schaich, Standardized methods for the
7. S. Gorinstein, M. Weisz, M. Zemser, K. Tilis, A. Stiller, I. Flam, determination of antioxidant capacity and phenolics in foods and
Y. Gat, Spectroscopic analysis of polyphenols in white wines. J. dietary supplements. J. Agric. Food. Chem. 53(10), 4290–4302
Ferment. Bioeng. 75(2), 115–120 (1993) (2005)
8. C. Henry, X. Vitrac, A. Decendit, R. Ennamany, S. Krisa, J.-M. 22. H. Schulz, M. Baranska, Identification and quantification of

Mérillon, Cellular uptake and efflux of trans-piceid and its agly- valuable plant substances by IR and Raman spectroscopy. Vib.
cone trans-resveratrol on the apical membrane of human intesti- Spectrosc. 43(1), 13–25 (2007)
nal Caco-2 cells. J. Agric. Food. Chem. 53(3), 798–803 (2005) 23. N. Sinelli, A. Spinardi, V. Di Egidio, I. Mignani, E. Casiraghi,
9. Y. Hu, Z.J. Pan, W. Liao, J. Li, P. Gruget, D.D. Kitts, X. Lu, Evaluation of quality and nutraceutical content of blueberries
Determination of antioxidant capacity and phenolic content of (Vaccinium corymbosum L.) by near and mid-infrared spectros-
chocolate by attenuated total reflectance-Fourier transformed- copy. Postharvest Biol. Technol. 50(1), 31–36 (2008)
infrared spectroscopy. Food. Chem. 202, 254–261 (2016) 24. Socrates G, Infrared and Raman Characteristic Group Frequen-
10. D. Huang, B. Ou, R.L. Prior, The chemistry behind antioxidant cies: Table and Charts, 3rd edn. (Wiley, Chichester, 2007)
capacity assays. J. Agric. Food. Chem. 53(6), 1841–1856 (2005) 25. A. Soriano, P. Pérez-Juan, A. Vicario, J. González, M. Pérez-
11. H.S. Lam, A. Proctor, L. Howard, M.J. Cho, Rapid fruit extracts Coello, Determination of anthocyanins in red wine using a newly
antioxidant capacity determination by Fourier transform infrared developed method based on Fourier transform infrared spectros-
spectroscopy. J. Food Sci. 70(9), C545–C549 (2005) copy. Food. Chem. 104(3), 1295–1303 (2007)
12. V. Lohachoompol, M. Mulholland, G. Srzednicki, J. Craske,
26. L.-Q. Sun, X.-P. Ding, J. Qi, H. Yu, He S-a, J. Zhang, H.-X. Ge,
Determination of anthocyanins in various cultivars of highbush B.-Y. Yu, Antioxidant anthocyanins screening through spec-
and rabbiteye blueberries. Food. Chem. 111(1), 249–254 (2008) trum–effect relationships and DPPH-HPLC-DAD analysis on
13. X. Lu, B.A. Rasco, Determination of antioxidant content and nine cultivars of introduced rabbiteye blueberry in China. Food.
antioxidant activity in foods using infrared spectroscopy and Chem. 132(2), 759–765 (2012)
chemometrics: a review. Crit. Rev. Food Sci. Nutr. 52(10), 853– 27. K. Thaipong, U. Boonprakob, K. Crosby, L. Cisneros-Zevallos,
875 (2012) D.H. Byrne, Comparison of ABTS, DPPH, FRAP, and ORAC
14. X. Lu, J. Wang, H.M. Al-Qadiri, C.F. Ross, J.R. Powers, J.
assays for estimating antioxidant activity from guava fruit
Tang, B.A. Rasco, Determination of total phenolic content and extracts. J. Food Compos. Anal. 19(6), 669–675 (2006)
antioxidant capacity of onion (Allium cepa) and shallot (Allium 28. S.Y. Wang, H. Chen, M.J. Camp, M.K. Ehlenfeldt, Genotype
oschaninii) using infrared spectroscopy. Food. Chem. 129(2), and growing season influence blueberry antioxidant capacity and
637–644 (2011) other quality attributes. Int. J. Food Sci. Technol. 47(7), 1540–
15. J.-C. Merlin, A. Statoua, J.-P. Cornard, M. Saidi-Idrissi, R.
1549 (2012)
Brouillard, The international journal of plant biochemistry reso- 29. M.H. Zhang, J. Luypaert, J.A. Fernández Pierna, Q.S. Xu, D.L.
nance Raman spectroscopic studies of anthocyanins and antho- Massart, Determination of total antioxidant capacity in green tea
cyanidins in aqueous solutions. Phytochemistry 35(1), 227–232 by near-infrared spectroscopy and multivariate calibration. Tal-
(1993) anta 62(1), 25–35 (2004)

13