Food Control 52 (2015) 103e111

Contents lists available at ScienceDirect

Food Control
journal homepage: www.elsevier.com/locate/foodcont

Mid-infrared spectroscopy for discrimination and classification of

Aspergillus spp. contamination in peanuts
Hande Kaya-Celiker a, P. Kumar Mallikarjunan a, *, Archileo Kaaya b
a
Biological Systems Engineering, Virginia Tech, Blacksburg, VA, USA
b
Archileo Kaaya, Food Technology and Human Nutrition, Makerere University, Kampala, Uganda

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

Article history: In this study, Aspergillus spp., common colonists in peanut, were characterized, classified and quantified
Received 24 September 2014 using FTIR coupled with ATR accessory. FTIR-ATR spectral data of infected peanut samples were pre-
Received in revised form processed (mean-centering, smoothing the 1st derivative), and used for the PLS regression analysis for
11 December 2014
quantitative results. Very high R2 values (96.20e99.98%) together with low error of RMSEC values (0.014
Accepted 14 December 2014
Available online 23 December 2014
e0.153 Log CFU/g of peanut) were obtained. Even, the spectrum of peanut matrix was dominant at early
stages of invasion (
2.5 Log CFU/g peanut), resulting in section separation (Nigri from Flavi) and at higher
population (>4 Log CFU/g, species level separation (Aspergillus alliaceus, Aspergillus caelatus, Aspergillus
Keywords:
FTIR-ATR
flavus, Aspergillus parasiticus, and Aspergillus tamari) was observed. The accuracy of correct classification
Peanut increased proportionally with fungal invasion level and 100% correct classification was reached when the
Aspergillus species cell level was Log CFU/g ¼ 4.5e5. Samples with similar secondary metabolites (toxin producers) grouped
PLS regression close-by in PC score diagrams for all levels of fungal growth. Results highlight the possible imple-
Discriminant analysis mentation of FTIR-ATR model to detect infected peanuts even at early stages of invasion; besides, to
prove the potential separation capability in terms of species and their secondary metabolites.
© 2014 Elsevier Ltd. All rights reserved.

1. Introduction Cabanes, 1994), which is considered as a potent carcinogen in rats

The genus Aspergillus is ubiquitous in nature and distributed
worldwide. It is among the most studied of all fungal genera due to
their economic impact as being both industrially important bio-
producer of certain enzymes and a causative agent in food
spoilage (Bennett, 2010). The genus contains about 250 species and
within the genus Aspergillus there are eight subgenera; and one or
more sections assigned to each subgenus (Raper & Fennell, 1965;
Samson & Varga, 2010). Currently, there are 18 sections named, in
total (Samson & Varga, 2010). Among them, section Flavi, which
takes its name from infamous member, Aspergillus flavus, and the
black aspergilli, Aspergillus niger have been frequently seen in
peanuts as dominant colonists. A. niger has always been important
for biotechnology applications; for example, it's been used for
synthesis of industrially valuable products like citric acid, or glu-
conic acid (Bennett, 2010). However, A. niger is also known for its
potential to produce ochratoxin (Abarca, Bragulat, Castella, &
* Corresponding author. Tel.: þ1 540 231 7937; fax: þ1 540 231 3199.
E-mail addresses: hande@vt.edu (H. Kaya-Celiker), kumar@vt.edu
(P.K. Mallikarjunan), ankaaya@agric.mak.ac.ug (A. Kaaya).
(Mantle, Kulinskaya, & Nestler, 2005). A. flavus together with
Aspergillus parasiticus are the most commonly seen opportunistic
pathogen of crops and occur regularly in all-inclusive food supplies
but especially seen in oil-rich seeds (Diener et al., 1987) like pea-
nuts. Peanut is one of the most vulnerable crops to Aspergillus spp.
contamination (Kaaya, Harris, & Eigel, 2006). Peanut pods are in
direct contact with soil, which is the source of primary inoculum for
Aspergillus spp., thus invasion of seed by these species before har-
vest is common (Horn, 2005). Other section Flavi species in peanut
that have been reported are Aspergillus caelatus (Horn, 1997) and A
tamari (Horn et al., 1996), which are not aflatoxin producers; but
Aspergillus tamari produces cyclopiazonic acid. Aspergillus alliaceus,
which have recently been placed in section Flavi (Peterson, 2008)
has also been reported as being an ochratoxin A producer (Bayman,
Baker, Doster, Michailides, & Mahoney, 2002) and may colonize on
peanut seeds depending on the environmental conditions of the
field (Horn, 2005).
Bearing the importance of mycotoxins in food and feed, there is
a need for rapid detection of toxic metabolites and their source
fungi; and many biochemical and molecular techniques are avail-
able today (Gilbert & Anklam, 2002). Fourier transform infrared
spectroscopy (FTIR) technology has gaining popularity in the area

http://dx.doi.org/10.1016/j.foodcont.2014.12.013
0956-7135/© 2014 Elsevier Ltd. All rights reserved.
104 H. Kaya-Celiker et al. / Food Control 52 (2015) 103e111
of identification and classification of microorganisms compared to
existing laborious and slow methods; and successful applications
were built up even in subspecies level in years. Many authors have
reported the use of mid-infrared spectral data and chemometrics as
powerful discrimination tool for microorganisms and have been
reviewed elsewhere (Mariey, Signolle, Amiel, & Travert, 2001). FTIR
in combination with reflectance accessorizes like attenuated total
reflectance (ATR) (Kos, Lohninger, & Krska, 2003) or photoacoustic
spectroscopy (PAS) (Gordon, Jones, McClelland, Wicklow, & Greene,
1999) have been proposed for Fusarium graminearum and A. flavus
detection on corn, respectively. Recently, a research group in Ger-
many has shown successful usage of FTIR to identify airborne fungi
and rapidly characterized “microfungi” stain (Fischer, Braun,
Thissen, & Dott, 2006). Fischer and colleagues have also demon-
strated different sample preparation protocols to get reproducible
procedure and repetitive spectral data of the living mycelia cells
(Fischer et al., 2006). In all these taxonomic classification studies,
including the bacterial cell identification (Naumann, 2000) quality
of the developed multivariate regression or classification models
depends on data compatibility, which was provided by standard-
izing the sample preparation procedures (Beekes, Lasch, &
Naumann, 2007). In this study, it was aimed to interpret spectral
characteristics of Aspergillus spp. growing on peanut. It was pro-
posed that different Aspergillus species have different growth pro-
file and different metabolism which can be detected by FTIR-ATR in
combination with multivariate analysis models. By doing so, sam-
ple preparation step was minimized (no need for isolation of fungi),
peanut matrix was used as bulk medium and species classification
was designed.
2. Methodology
2.1. Aspergillus spp. spore solution preparation
Eight different strains, namely: A. alliaceus NRRL 4181,
A. caelatus NRRL 25528, A. flavus NRRL 1957 (non aflatoxin pro-
ducing mutant), A. flavus NRRL 3357 (aflatoxin producer),
A. parasiticus NRRL 21369 (non aflatoxin producing UV color
mutant), A. parasiticus NRRL 5862 (aflatoxin producer), A. tamari
NRRL 20818 and A. niger NRRL 326 were kindly provided by Dr.
Bruce Horn (USDA National peanut Research Lab., Georgia, USA).
Conidia of each strain from stock cultures were further inoculated
on potato dextrose agar and incubated at room temperature for 7
days to enable significant sporulation to take place. After incuba-
tion, 5 ml of sterile distilled water was aseptically added to each
plate. A sterile plastic inoculation loop was used to loosen the
colonies from the PDA plates. The suspension created was then
filtered through sterile cheese cloth into a sterile 50 ml capacity
Falcon tube.
2.2. Infecting peanut seeds with Aspergillus spp.
Smooth, blanched, Virginia peanuts were obtained from local
retailers and stored at 4  C in reclosable plastic bags until analysis.
Peanut pods were surface sterilized before inoculation treatment
using 1.0% NaOCl for 3 min and rinsed thoroughly with distilled
water. After removing the excess water with pre-sterilized paper
towel, each pod was placed into petri-dish moist chamber (5 pods
per chamber, 2 chambers for each species), consisting of moist filter
paper to ensure high humidity (up to 80e100%). 10 ml spore sus-
pensions (105e106 CFU/ml) were dispersed onto surface sterilized
peanut seeds and were incubated in chamber at room temperature
(23 ± 1  C) for 3 and 5 days to render contaminated peanuts at
different levels of mold concentrations. Mold growth was visible
after 2 days. Seeds covered with masses of fungi were mixed with
pre-sterilized peanuts to obtain samples having varying cell con-
centrations of Aspergillus spp. of interest that will function as cali-
bration set. Paste was made for homogenous mold distribution by
grinding raw peanuts with a food processor (Butterfly Emerald
Mixer, Gandhimathi Appliances Ltd., Tamil Nadu, India) equipped
with metal cutting blade and stainless steel container. Initial fungal
load was determined in terms of colony plate count and results
were expressed as averages Log CFU/g of peanuts of three mea-
surements. Samples were kept in sealed containers and stored at
4  C until analysis.
2.3. FTIR-ATR spectral measurements
The infrared spectra of contaminated peanut paste samples
were collected using FTIR spectrometer (Nicholet 6700, Thermo
Fisher Scientific Inc., Madison, WI, USA) coupled with DTGS KBr
detector. Sample compartment was equipped with Smart iTR dia-
mond attenuated total reflectance (ATR) accessory (Thermo Fisher
Scientific Inc., Madison, WI, USA) consisting of laminated diamond
as crystal material, mounted in stainless-steel plate. The diamond
crystal has a reflective index of 2.4 at 1000 cm1 and an angle of
incidence of 45 . Spectra were attained and processed with the
Omnic Spectra (Thermo Scientific) software. Spectra were scanned
over a range of 4000e625 cm1 at a resolution of 4 cm1 and a gain
of 2.0. Average of 64 scans was used as final spectrum. Fourier
transformation was performed with Mertz phase correction, NeB
strong apodization function, with a zero filling factor of 2. The
crystal was cleaned between successive measurements with 70%
methanol and dried truly. The cleaned crystal was checked to avoid
residues from previous measurements by observing the back-
ground spectrum. Three repetitive spectral readings were collected
per sample and to include the blocking effect, few of randomly
selected samples were run again, by analyzing some of the samples
on the following day. All of the collected spectra were used for
analysis, not the averages. Samples prepared freshly (only stored
during the course of the FTIR analysis) were spread onto diamond
crystal until full contact was provided. Pressure knob was not used
to avoid expressing oil content of paste upon pressure. Refrigerated
samples were left at room temperature (23 ± 1  C) for 2 h to let
samples reach to room temperature before analysis.
2.4. Multivariate data analysis
2.4.1. PLS regression
Partial least square (PLS) modeling is a powerful multivariate
analysis tool for spectral analysis of samples having complex
structure as is the case in peanut and mold itself. In this study,
peanuts were infected with different strains of genus Aspergillus
and kept under high humid condition at room temperature to
accelerate the growth of fungi; and the spectral responses were
analyzed with PLS regression technique (TurboQuant IR calibration
and prediction software, Thermo Fisher). Contaminated peanut
paste samples had cell concentration values varied between 2.5 and
6.0 Log CFU/g peanut. Readouts collected for each A. spp. were
40e50 for calibration set and 20e24 for prediction set. Differences
in the number of readouts were because some samples have more
readings to include blocking effect. Spectral readings were pre-
processed before loaded into PLS regression in order to reduce
the variability associated with unrelated sources affecting the in-
tensity of all peaks. Data were mean centered to remove the de-
pendency on magnitude and filtered to smooth the peaks that
resemble random noise. First derivative of the data was filtered by
Savitzky-Golay smoothing filter (data points ¼ 7 and polynomial
order ¼ 3).
 H. Kaya-Celiker et al. / Food Control 52 (2015) 103e111 105

The performance of developed PLS regression models for each
Aspergillus spp. were tested by applying “leave one out” cross-
validation method in which each calibration standard was quanti-
fied as validation standard. The performance of developed PLS
models were evaluated through correlation coefficient of deter-
mination (R2), root mean-squared error of calibration (RMSEC), root
mean-squared error of cross-validation (RMSEV) and root mean-
squared error of prediction (RMSEP).
2.4.2. Discriminant analysis
In order to determine the threshold cell concentration value
which results in full discrimination of all Aspergillus spp. used in
this study, samples were separated in four batches according to
fungal growth level as samples having cell concentration of
2.5 Log CFU/g peanut and lower, 3.0 to 4.0 Log CFU/g peanut, 4.5 to
5 Log CFU/g peanut, and 5.5 Log CFU/g peanut and higher. Pre-
processed (mean-centered, filtered by smoothing the 1st deriva-
tive) spectral data of each batch was evaluated with respect to
quality of the classification using Discriminant analysis technique
(TurboQuant IR calibration and prediction software, Thermo Fisher)
by computing the distance from each class center in Mahalanobis
distance units. During calibration, the software computes a mean
spectrum and then generates a distribution model by estimating
the variance at each frequency in the analysis range. Results were
presented as principal component (PC) score plots.
3. Results and discussion
Peanut seeds were aseptically inoculated with spore solutions of
Aspergillus spp. and kept at humid environment until the masses of
fungi of interest covered the seeds. Fig. 1 shows the average of 5 re-
petitive reading of fungal mycelium of Aspergillus spp. scraped off the
peanut surfaces after drying infected peanuts in desiccator cabinet
with drierite. There are minor spectral differences within the species
of genus Aspergillus, but they all predominantly give intensive ab-
sorption bands between 3600e3100 cm1 and 1700e1500 cm1
which are assigned to protein content (amide A, amide I and amide II
bands) of the fungi. If we briefly interpret the spectral profile, ab-
sorption bands observed between 3050e2750 cm1 and
1500e1300 cm1 are generally led by lipid content but there are also
some deformation modes of vibrations of protein structures. The re-
gion between 1200 and 900 cm1 is referred as ring vibrations of
oligo- and polysaccharides (Naumann, 2000) and region between 900
and 600 cm1 is generally arising from aromatic ring vibrations (Fig.1).
Since the early 1950s, there have been abundant of taxonomic
classification works published in the literature investigating
different spectral windows to get optimum dispersion of strains of
different bacterial or fungal cells (for review of early studies see
(Mariey et al., 2001). Among many advantages of usage of infrared
spectroscopy, analysis is limited to microorganisms that can grow
in culture media and stable and strictly controlled cultivation
conditions (Naumann, 2000). Standardization of isolation of pure
cultures from food matrix is one of the most time consuming step
for discrimination studies. When the fungal mycelium of Aspergillus
spp. was loaded to Discriminant analysis tool, at some extent,
separation of different species were observed (Fig. 2). In this
analysis the spectral range of whole spectrum, 4000e650 cm1,
was used. Improving the model and standardizing the isolation
methodology, better discrimination of each species can be ach-
ieved. But, in this study, it was aimed to describe methodology that
reduces the sample preparation needs and do classification using
the peanut matrix as it is; also, to depict the biochemical alterations
in both cell structure and nutritious medium (peanuts) as fungal
growth happens. Thus, spectral profiles of Aspergillus species (Fig. 1)
and the variation and correlation spectrum obtained from the PLS
regression analysis (Fig. 3) were all used for spectral window

Mean spectrum of A.alliaceus NRRL 4181

Abs

0.2

0.4 Mean spectrum of A.caelatus NRRL 25528

Abs

0.2

0.4 Mean spectrum of A.flavus NRRL 1957

Abs

0.2

Mean spectrum of A.flavus NRRL 3357

Abs

0.2

0.4 Mean spectrum of A.niger NRRL 326

Abs

0.2

Mean spectrum of A.parasiticus NRRL 5862

Abs

0.2

Mean spectrum of A.parasiticus NRRL 21369

Abs

0.2

0.2 Mean spectrum of A.tamarii NRRL 20818

Abs

400 0 350 0 300 0 250 0 200 0 150 0 100 0

W a venum bers (c m -1)

Fig. 1. Mean spectra of each Aspergillus spp. studied. Mycelium of fungi of interest was scrapped off peanut surface, dried and loaded onto diamond cell of ATR unit for spectrum
collection.
106 H. Kaya-Celiker et al. / Food Control 52 (2015) 103e111

Table 1
Functional groups and vibration modes of FTIR spectral regions of fingerprint region
(Guillen & Cabo, 1997; Irudayaraj et al., 2001; Ismail et al., 1993).

Frequency (cm1) Functional group Mode of vibration

1743 eC]O (ester) Stretching

1710 eC]O (acid) Stretching
1643 C]O, CeN Stretching
1544 NeH, CeO Bending
CeC, CeN Stretching
1464 eCeH (CH2, CH3) Bending (scissoring)
1416 ]CeH (cis) Bending (rocking)
1378 eCeH (CH3) Bending (sym)
1238 eCeH (CH3) Bending (sym)
1160 eCeO, eCH2e Stretching, bending
1118 eCeO, eCH2e Stretching, bending
1095 eCeO Stretching

and the region representing the absorptions from the diamond

Fig. 2. Principal Component (PC) score plot for classes of different Aspergillus spp.
studied. ( ) A. niger NRRL 326; ( ) A. flavus NRRL 3357; ( ) A. flavus NRRL 1957; ( )
A. parasiticus NRRL 5862; ( ) A. parasiticus NRRL 21369; (◊) A. alliaceus NRRL 4181; (þ)
A. caelatus NRLL 25528; ( ) A. tamari NRRL 20818.
selection and interpretation of peaks. The variation and correlation
spectra of each species of genus Aspergillus displayed regions that
are active and the association of these regions with respect to cell
concentration, respectively (Fig. 3).
3.1. Spectral region selection and interpretation
Given the complex structure of both Aspergillus spp. and peanut
matrices, the spectra can contain a superposition of hundreds of
infrared modes. The OeH stretching of water (3600e3200 cm1)
crystal (2340e1800 cm1) were excluded from all model calibra-
tions. Wave numbers between 3050 and 2750 cm1 were also
excluded from data analysis and the active portion of the spectrum
that lies between 1800 and 650 cm1 (fingerprint region) were
used in the analysis. This region covers the variations that are un-
ambiguous to fungal growth (Kos, Lohninger, & Krska, 2002).
Detailed band assignments of selected peaks in specified region
were summarized in Table 1. Representative spectral regions of
fungal growth on peanut were further illustrated in Fig. 4a, b and c,
and area alterations with respect to increased Log CFU/g peanut
values were indicated with arrows. Peanuts are complex structures,
with the main components being water (very little), proteins, fats
and carbohydrates. Proteins appear as amide I (1700e1600 cm1)
and amide II (1565e1520 cm1) groups (Stuart, 2004). The protein
related peaks were assigned to 1643 cm1 and 1544 cm1 in the
current study. The first one represents the carbonyl and CeN
stretching of the amide I group and the second one relates to

Fig. 3. PLS regression resulting statistical spectra: a: overall variance spectrum; b: multiple correlation spectrum; c: mean spectrum of infected peanut samples with A. flavus NRRL
1957 and d: mean spectrum of clean peanut samples.
 H. Kaya-Celiker et al. / Food Control 52 (2015) 103e111 107

Fig. 4. Spectral changes as A. flavus NRRL 1957 grows on peanut. Arrows show the direction of increasing A. flavus NRRL 1957 concentration (Log CFU/g peanut).

combination of the CeC and CeN stretching, and NeH and CeO
bending vibrations of the amide II group (Irudayaraj, Sivakesava,
Kamath, & Yang, 2001). Another protein related band is usually
observed around 3290 cm1 as a broad peak (NeH stretching of
amide group) but this region was not included in regression anal-
ysis. As stated before, fungal infection can be determined from the
increased absorption peaks of proteins, and decreased absorptions
at lipid and carbohydrate regions due to the metabolic activity of
fungus (Naumann, 2000). In all Aspergillus spp. profiles, a sharp
decrease and a steady increase after cell concentration reaches to
5 Log CFU/g of peanut was observed when amide I and amide II
band components were investigated. Same trend was noted for
amide A band (3290 cm1) but depth of concavity was smaller. It's
been reported that proteolytic enzymes of fungi can hydrolyze the
host proteins and can convert the building blocks into fungal pro-
tein during advanced stages of deterioration (Bothast, 1978). There
is also a host-fungus interrelationship, in which the level of protein,
protease activity or amount of free amino acids differ significantly
among the species of Aspergillus grown on peanut (Achar,
Sreenivasa, & Galdo, 2011); and only when heavily infected, free
amino acid content shows a trend with respect to fungal growth
level (Chiou, 1997). Similarly, our results suggest fungal metabolism
108 H. Kaya-Celiker et al. / Food Control 52 (2015) 103e111
of peanut proteins were metabolized at first (concave down), and
then, fungal proteins started to be spectrally observable after a
certain amount of fungal growth (concave up).
Along with protein peaks, the FTIR spectra of infected and clean
peanut paste samples also showed a sharp linear increase and/or
decrease at fat associated peaks indicating lipid hydrolysis. Peanut
is composed of mixed glycerides and contains several fatty acids,
accounting for its high proportion of unsaturated structure (almost
80% oleic and linoleic acid content (Caballero, Trugo, & Finglas,
2003); this structure is known to enhance the growth rate of
Aspergillus spp. (Fanelli & Fabbri, 1980). Beside the growth rate, the
percentage of polyunsaturated fatty acids in peanuts affects the
amount of mycotoxin produced (e.g., results in induced aflatoxin
production by lipoperoxides in A. parasiticus and A. flavus (Fabbri,
Fanelli, Panfili, Passi, & Fasella, 1983)). The enzymatic degradation
of lipids into free fatty acids and glycerol has been shown to be
prior to metabolism of starch granules or protein content (Smart,
Wicklow, & Caldwell, 1990), during Aspergillus invasion; alterna-
tively, in some studies drop in sugar concentration was reported to
be before or simultaneous to triglyceride hydrolysis through
metabolism of Aspergillus (Mellon, Cotty, & Dowd, 2000; Mellon,
Dowd, & Cotty, 2002). Still, hydrolysis of triglycerides by the ac-
tion of lipases has been an important alert to mold growth and
consequently increase in the free fatty acid content of grain has
been suggested as an indicator of grain deterioration (Bothast,
1978). The peak height of the fatty acid ester linkage (eC]O)
centering at 1743 cm1 decreased due to lipid consumption and
simultaneously ester carbonyl (eC]O stretching of acid) of
breakdown product of free fatty acids (FFA) appeared and steadily
increased at 1710 cm1 (Ismail, Vandevoort, Emo, & Sedman, 1993).
Another fat associated band of CeO stretching at 1160 cme1
generally decreased as mold count increased. These bands can be
used for determination of level of Aspergillus spp. contamination in
peanut samples (Fig. 5), because all three gave a good correlation
with high coefficient of determination, R2 values of 98.21, 98.71,
99.37% for A. flavus NRRL 1957 (as an example), respectively, when
linearly regressed to mold amount (similar trends were observed
for other species, too). Apart from these peaks, other fat related
peaks at 1464 and 1378 cme1, which represent bending vibrations
of the methylene and symmetrical bending vibration of methyl
groups of the fatty acids, respectively, showed a slight drop in
absorbance until mold counts between 20,000 and 100,000 cells;
afterward increased to initial absorbance value as invasion
advanced. The height of peak at 1238 cm1, which related to the
CeO stretching of esters, had a similar trend, whereas the band at
0.30
0.25
Absorbance
0.20
0.15
0.10
0.05
0.00
0 100000 200000 300000
Mold count (CFU/g)
Fig. 5. Average of 3 absorbance readings with respect to cell count (CFU/g peanut) for
▫
A. flavus NRRL 1957 at specified wave numbers: (◊) at 1743 cm1; ( ) at 1710 cm1;
(B) at 1160 cm1. The corresponding R2 values and the equations of fitting curves are:
▪
(dd) y ¼ 2*107x þ 0.2408 & R2 ¼ 0.9848; (d d) y ¼ 2*107x þ 0.02 &
R2 ¼ 0.9879; (- - -) y ¼ 1*107x þ 0.1798 & R2 ¼ 0.9935.
1118 and 1095 cm1 decreased with increased mold growth. Those
bands may be assigned to CeO group in esters (Guillen & Cabo,
1997) of triglycerides and can be used for estimating the level of
lipid degradation or fungal growth. Differently, the peak at
1416 cm1 rose with half of the rate compared to previous bands in
regard to Aspergillus count.
In our earlier studies, we tried to spike the peanut paste with
Aspergillus spore solution. However, no detectable, visible change in
spectrum was observed (data not shown), and peanut was domi-
nant in all aspects. Current results prove that metabolic activity of
fungus is inevitably the leading factor that affects the spectral
readings after certain level of infection is reached. Besides, different
species have different metabolism, different primary or secondary
metabolites, thus, different chemical profiles at different levels of
growth. These chemical alterations were the basis of Discriminant
analysis and classification with respect to mold level.
3.2. PLS model calibration and validation
PLS regression analysis is to predict dependent variables from a
set of independent variables achieved by extracting a set of latent
variables which have the best predictive power and to obtain a
relationship between the FTIR predicted and actual values. Separate
PLS regression models were developed for each species including
the cell concentration information in terms of Log CFU/g of peanut
by deciding on the minimum number of factors accounting for as
much of the manifest variation as possible while modeling the
responses well. Pre-processed data (mean-centered, filtered by
smoothing the 1st derivative) was loaded into PLS (SIMPLE algo-
rithm) and resulted regression parameters were summarized in
Table 2. All models fit the actual cell concentration values well with
R2 values ranging between 96.20% (A. alliaceus NRRL 4181) and
99.98% (A. flavus NRRL 1957); and the RMSEC values were between
0.153 and 0.014 Log CFU/g of peanut for the same species, respec-
tively. The highest RMSEP, which is an indicator of the accuracy of
the model, was observed for A. tamari NRRL 20818 with a value of
0.235 Log CFU/g of peanut. When % difference was investigated, it
was higher for prediction standards than the data matrices used for
calibration, which is probably a result of large factors number used.
This indicates an over-fit of calibration model developed for
A. tamari NRRL 20818. Similarly the highest RMSEV was found for
A. tamari NRRL 20818 (0.280 Log CFU/g of peanut), while A. niger
NRRL 326 gave the lowest errors with values of 0.153 and
0.120 Log CFU/g of peanut for RMSEV and RMSEP, respectively. The
factor numbers were decided according to the predicted residual
error sum of squares (PRESS) plot. PRESS value decreases as the
factors are added. At some point error value reaches down to a
minimum, which gives the number of factors used in PLS. Adding
more factors will result in over-fitted calibration models. The
number of factors used for each species of Aspergillus was given in
Table 2 and actual versus predicted cell count values were illus-
trated in Fig. 6. In general, low levels of errors of calibration, vali-
dation and prediction data sets for all species studied prove the
high performance of FTIR when used as an analytical tool for esti-
mation of mold count. In other words, level of infection by Asper-
gillus spp. on peanuts can be determined with very little sample
preparation and without usage of chemicals in a very short time
(less than 1 min.).
400000
3.3. Classification with respect to cell concentration
Firstly, the whole data pool was separated into 4 batches (each
having 8 classes representing different Aspergillus spp.) and sepa-
rate Discriminant Analyses were applied to investigate the
threshold cell count values to get full discrimination at species
 H. Kaya-Celiker et al. / Food Control 52 (2015) 103e111 109

Table 2
PLS regression statistics of each Aspergillus spp. studied.

Index Aspergillus spp. studied Factor Slope Intercept R2 (%) RMSEC (Log CFU/g) RMSECV (Log CFU/g) RMSEP (Log CFU/g)

1 A. niger NRRL 326 6 1.006 0.032 99.55 0.077 0.153 0.120

2 A. caelatus NRRL 25528 6 0.990 0.041 99.55 0.077 0.223 0.140
3 A. tamari NRRL 20818 9 0.983 0.045 99.90 0.035 0.280 0.235
4 A. flavus NRRL 1957 9 0.993 0.031 99.98 0.014 0.153 0.169
5 A. flavus NRRL 3357 5 0.977 0.098 98.96 0.117 0.246 0.189
6 A. parasiticus NRRL 5862 6 0.989 0.043 99.67 0.068 0.197 0.164
7 A. parasiticus NRRL 21369 6 0.994 0.034 99.65 0.068 0.1160 0.139
8 A. alliaceus NRRL 4181 3 0.972 0.094 96.20 0.153 0.244 0.195

6
FTIR Predicted Cell Count Values (Log CFU/g)

1
1 2 3 4 5 6
Actual Cell Count Values (LogCFU/g)
▵
Fig. 6. FTIR predicted versus actual cell count (Log CFU\g) values for all Aspergillus species studied: ( ) A. niger NRRL 326; ( ) A. flavus NRRL 3357; ( ) A. flavus NRRL 1957; ( )
A. parasiticus NRRL 5862; ( ) A. parasiticus NRRL 21369; ( ) A. alliaceus NRRL 4181; ( )A. caelatus NRLL 25528; ( ) A. tamari NRRL 20818.

level. Ten principal components were used to calibrate the method
for all data sets; and 74.1, 76.5, 77.6, and 79.6 percent variability
were described for groups of Log CFU/g
2.5, Log CFU/g ¼ 3.0e4.0,
Log CFU/g ¼ 4.5e5, and Log CFU/g  5.5, respectively.
The PC scores diagnostics for each group were illustrated in
Fig. 7aed. PC scores of each class (Aspergillus spp.) had a tendency
to depart from other classes as the cell concentration increases, as
expected. When the cell counts were lower than 500, all species
clustered close to each other, still a separation could be observed
(Fig. 7a). Within the group of 2.5 Log CFU/g peanut and lower, 6
standards were misclassified; and misclassified sample number
decreased to 3 for the group of 3.0e4.0 Log CFU/g peanut. For low
level of Aspergillus invasion, it is obvious that the spectral changes
were predominantly from peanut matrix. But, after mold count of
1000 CFU/g was reached, first the A. niger NRRL 326 separated
which was followed by A. tamari NRRL 20818 and UV color mutant
A. parasiticus NRRL 21369 (Fig. 7b). In groups of Log CFU/g ¼ 4.5e5,
and LogCFU/g  5.5 all sample data belongs to 8 species of genus
Aspergillus were classified correctly (Fig. 7c and Fig. 7d). The most
conspicuous observation was that for all cell concentration values
the class centers of A. flavus 3357 and A. parasiticus NRRL 5862
remained close by. These strains are known to be an aflatoxin
producer, and apparently similar secondary metabolites produced
increases the similarity within the spectral readouts. Previously,
different metabolites produced by strains in the same taxon were
reported to be discriminated well (Fischer et al., 2006). Likewise,
the level and type of mycotoxin produced is related to the level of
infection, and this was used as an indicator for discrimination and
quantification of different Aspergillus spp. with respect to their
secondary metabolites (Kaya-Celiker, Mallikarjunan, Schmale, &
Christie, 2014). This can be a good explanation for A. niger
belonging to section Nigri getting separated first, even the cell
count was around 1000 CFU/g. Also, A. niger is known to produce
ochratoxin A. All other 7 species were from section Flavi, and strain
level separation required more cell count (around 10,000) to be
fully distinguished.
4. Conclusion
Both peanut matrix and the invaded fungi have complex
structures, and both are readily giving absorptions at similar fre-
quencies, or masking the spectral contributions of one another. At
low level of fungal growth (Log CFU/g
2.5), peanut was found to
be main component in spectral readouts. However, some compo-
sitional changes like a decrease and consequent increase at protein
related peaks, or steady drop in lipid associated peaks indicated
that fungus of interest became dominant in the medium. PLS
regression results prove the potential of FTIR-ATR as a tool for
110 H. Kaya-Celiker et al. / Food Control 52 (2015) 103e111
Fig. 7. PC score plot for groups: (a) Log CFU/g
2.5; (b) Log CFU/g ¼ 3.0e4.0; (c) Log CFU/g ¼ 4.5e5; (d) Log CFU/g  5.5. Aspergillus species are: ( ) A. niger NRRL 326; ( ) A. flavus
NRRL 3357; ( ) A. flavus NRRL 1957; ( ) A. parasiticus NRRL 5862; ( ) A. parasiticus NRRL 21369; (◊) A. alliaceus NRRL 4181; (þ) A. caelatus NRLL 25528; ( ) A. tamari NRRL 20818.
quantitative analysis of fungal growth on food matrix because very
high R2 values together with low error percentages were found.
Besides, metabolic activity products of fungi added to spectral
readings and it was observed that fungi in the same section were
grouped together or split-up according to their secondary metab-
olites. This indicates that peanut kernels can be separated auto-
matically as healthy or invaded. Furthermore, according to the
stages of deterioration by fungi, level of secondary metabolites may
be decided.
This project is funded by Peanut CRSP (USAID) aiming to
improve livelihood of poor rural communities in East Africa by
emphasizing on health and nutrition aspects of aflatoxin and
Aspergillus infestation on peanuts. In many African countries,
especially in Uganda (where the project regionally focused on),
there is a serious lack of trained personnel/scientist and up-to-date
technology to manage and analyze these contaminants on peanuts.
FTIR-ATR based technology is capable of rapid and non-destructive
measures of Aspergillus spp. very easily (analysis of one sample
takes less than 1 min) and without the need for any harsh chemical
extractions. The method is not labor intensive as well. Potential
implementation of proposed model can improve the quality of
peanuts supplied by means of sorting out the infected peanuts; and
subsequently, conquer the health and trade constraint resulting
from fungal contamination.
Acknowledgments
Authors wish to express their gratitude to Peanut CRSP program of
USAID for providing the financial support to the project. Authors also
would like to extend their gratitude to Drs. Maria Elisa Christie, Justin
Barone and Charity Mutegi for their participation in the project.
References
Abarca, M. L., Bragulat, M. R., Castella, G., & Cabanes, F. J. (1994). Ochratoxin-A
production by strains of Aspergillus.niger var Niger. Applied and Environmental
Microbiology, 60(7), 2650e2652.
Achar, P. N., Sreenivasa, M. Y., & Galdo, G. (2011). Comparative studies on the
changes of total soluble proteins and protease activity in Georgian commercial
peanuts contaminated by Aspergillus flavus. Archives of Phytopathology and Plant
Protection, 45(2), 220e227. http://dx.doi.org/10.1080/03235408.2010.544465.
Bayman, P., Baker, J. L., Doster, M. A., Michailides, T. J., & Mahoney, N. E. (2002).
Ochratoxin production by the Aspergillus ochraceus group and Aspergillus
alliaceus. Applied and Environmental Microbiology, 68(5), 2326e2329.
Beekes, M., Lasch, P., & Naumann, D. (2007). Analytical applications of Fourier transform-
infrared (FT-IR) spectroscopy in microbiology and prion research. Veterinary
Microbiology, 123(4), 305e319. http://dx.doi.org/10.1016/j.vetmic.2007.04.010.
Bennett, J. W. (Ed.). (2010). An overview of the genus Aspergillus. Caister Academic.
Bothast, R. J. (Ed.). (1978). Fungal deterioration and related phenomena in cereals,
legumes and oilseeds. Westport, CT, USA: Food and Nutrition Press, Inc.
Caballero, B., Trugo, L. C., & Finglas, P. M. (2003). Encyclopedia of food sciences and
nutrition. Amsterdam: Academic Press.
Chiou, R. Y. Y. (1997). Estimation of fungal infection of peanut kernels by deter-
mination of free glutamic acid content. Applied and Environmental Microbiology,
63(3), 1083e1087.
Diener, U. L., Cole, R. J., Sanders, T. H., Payne, G. A., Lee, L. S., & Klich, M. A. (1987).
Epidemiology of aflatoxin formation by Aspergillus-flavus [Review] Annual Re-
view of Phytopathology, 25, 249e270. http://dx.doi.org/10.1146/
annurev.py.25.090187.001341.
Fabbri, A. A., Fanelli, C., Panfili, G., Passi, S., & Fasella, P. (1983). Lipoperoxidation and
aflatoxin biosynthesis by Aspergillus parasiticus and A. flavus. Journal of General
Microbiology, 129(NOV), 3447e3452.
Fanelli, C., & Fabbri, A. A. (1980). Growth requirements and lipid-metabolism of
Aspergillus-flavus [Article] Transactions of the British Mycological Society,
75(DEC), 371e375.
Fischer, G., Braun, S., Thissen, R., & Dott, W. (2006). FT-IR spectroscopy as a tool for
rapid identification and intra-species characterization of airborne filamentous
fungi [Article] Journal of Microbiological Methods, 64(1), 63e77. http://
dx.doi.org/10.1016/j.mimet.2005.04.005.
Gilbert, J., & Anklam, E. (2002). Validation of analytical methods for determining
mycotoxins in foodstuffs [Article] Trac-Trends in Analytical Chemistry, 21(6e7),
468e486. http://dx.doi.org/10.1016/s0165-9936(02)00604-0.
Gordon, S. H., Jones, R. W., McClelland, J. F., Wicklow, D. T., & Greene, R. V. (1999).
Transient infrared spectroscopy for detection of toxigenic fungus in corn: po-
tential for on-line evaluation. Journal of Agricultural and Food Chemistry, 47(12),
5267e5272. http://dx.doi.org/10.1021/jf990011f.
Guillen, M. D., & Cabo, N. (1997). Characterization of edible oils and lard by Fourier
transform infrared spectroscopy. Relationships between composition and frequency
of concrete bands in the fingerprint region. Journal of the American Oil Chemists
Society, 74(10), 1281e1286. http://dx.doi.org/10.1007/s11746-997-0058-4.
Horn, B. W. (1997). Aspergillus caelatus, a new species in section Flavi. Mycotaxon, 61,
185e191.
Horn, B. W. (2005). Colonization of wounded peanut seeds by soil fungi: selectivity
for species from Aspergillus section Flavi. Mycologia, 97(1), 202e217. http://
dx.doi.org/10.3852/mycologia.97.1.202.
 H. Kaya-Celiker et al. / Food Control 52 (2015) 103e111
Horn, B. W., Greene, R. L., Sobolev, V. S., Dorner, J. W., Powell, J. H., & Layton, R. C.
(1996). Association of morphology and mycotoxin production with vegetative
compatibility groups in Aspergillus flavus, A.parasiticus, and A. tamarii. Mycolo-
gia, 88(4), 574e587. http://dx.doi.org/10.2307/3761151.
Irudayaraj, J., Sivakesava, S., Kamath, S., & Yang, H. (2001). Monitoring chemical
changes in some foods using Fourier transform photoacoustic spectroscopy.
Journal of Food Science, 66(9), 1416e1421.
Ismail, A. A., Vandevoort, F. R., Emo, G., & Sedman, J. (1993). Rapid quantitative-
determination of free fatty-acids in fats and oils by Fourier-transform
infrared-spectroscopy. Journal of the American Oil Chemists Society, 70(4),
335e341. http://dx.doi.org/10.1007/bf02552703.
Kaaya, A. N., Harris, C., & Eigel, W. (2006). Peanut aflatoxin levels on farms and in
markets of Uganda. Peanut Science, 33(1), 68e75.
Kaya-Celiker, H., Mallikarjunan, P. K., Schmale, D., & Christie, M. E. (2014).
Discrimination of moldy peanuts with reference to aflatoxin using FTIR-ATR
system. Food Control, 44, 64e71. http://dx.doi.org/10.1016/
j.foodcont.2014.03.045.
Kos, G., Lohninger, H., & Krska, R. (2002). Fourier transform mid-infrared spec-
troscopy with attenuated total reflection (FT-IR/ATR) as a tool for the detection
of Fusarium fungi on maize. Vibrational Spectroscopy, 29(1e2), 115e119.
Kos, G., Lohninger, H., & Krska, R. (2003). Development of a method for the deter-
mination of Fusarium fungi on corn using mid-infrared spectroscopy with
attenuated total reflection and chemometrics. Analytical Chemistry, 75(5),
1211e1217. http://dx.doi.org/10.1021/ac0260903.
Mantle, P., Kulinskaya, E., & Nestler, S. (2005). Renal tumourigenesis in male rats in
response to chronic dietary ochratoxin A. Food Additives and Contaminants Part
a-Chemistry Analysis Control Exposure & Risk Assessment, 22, 58e64. http://
dx.doi.org/10.1080/02652030500358431.
111
Mariey, L., Signolle, J. P., Amiel, C., & Travert, J. (2001). Discrimination, classification,
identification of microorganisms using FTIR spectroscopy and chemometrics
[Review] Vibrational Spectroscopy, 26(2), 151e159. http://dx.doi.org/10.1016/
s0924-2031(01)00113-8.
Mellon, J. E., Cotty, P. J., & Dowd, M. K. (2000). Influence of lipids with and without
other cottonseed reserve materials on aflatoxin B-1 production by Aspergillus
flavus. Journal of Agricultural and Food Chemistry, 48(8), 3611e3615. http://
dx.doi.org/10.1021/jf0000878.
Mellon, J. E., Dowd, M. K., & Cotty, P. J. (2002). Time course study of substrate uti-
lization by Aspergillus flavus in medium simulating corn (Zea mays) kernels.
Journal of Agricultural and Food Chemistry, 50(3), 648e652. http://dx.doi.org/
10.1021/jf011048e.
Naumann, D. (2000). Infrared spectroscopy in microbiology. In R. A. Meyers (Ed.),
Encyclopedia of analytical chemistry: Applications, theory, and instrumentation.
Chichester, New York: Wiley.
Peterson, S. W. (2008). Phylogenetic analysis of Aspergillus species using DNA se-
quences from four loci. Mycologia, 100(2), 205e226. http://dx.doi.org/10.3852/
mycologia.100.2.205.
Raper, K. B., & Fennell, D. I. (1965). The genus Aspergillus. Baltimore: Williams and
Wilkins.
Samson, R. A., & Varga, J. (Eds.). (2010). An overview of the genus Aspergillus.
Wymondham, Norfolk, UK: Caister Academic.
Smart, M. G., Wicklow, D. T., & Caldwell, R. W. (1990). Pathogenesis in Aspergillus ear
rot of maize e light-microscopy of fungal spread from wounds. Phytopathology,
80(12), 1287e1294. http://dx.doi.org/10.1094/Phyto-80-1287.
Stuart, B. H. (2004). Infrared spectroscopy: Fundamentals and applications. John Wiley
& Sons Ltd.