Spectrochimica Acta Part B 141 (2018) 70–79

Contents lists available at ScienceDirect

Spectrochimica Acta Part B

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

Localization and distribution of Zn and Fe in grains of biofortified

bread wheat lines through micro- and triaxial-X-ray
fluorescence spectrometry夽
P. Cardoso a , T.C. Mateus a , G. Velu b , R.P. Singh b , J.P. Santos a , M.L. Carvalho a ,
V.M. Lourenço c , F. Lidon d , F. Reboredo d , M. Guerra a, *
a
Laboratório de Instrumentação, Engenharia Biomédica e Física da Radiação (LIBPhys-UNL), Departamento de Física, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa,
Monte da Caparica, Caparica 2892-516, Portugal
b
International Maize and Wheat Improvement Center (CIMMYT), Apdo postal 6-641, Mexico DF, Mexico
c
GeoBioTec & Departamento de Matemática, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, Caparica 2829-516, Portugal
d
GeoBioTec, Departamento de Ciências da Terra, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, Caparica 2829-516, Portugal

A R T I C L E I N F O A B S T R A C T

Article history: X-ray fluorescence analysis has been performed in wheat grains from a field trial where some biofortified
Received 17 October 2017 and non-biofortified wheat varieties were subjected to Zn biofortification through soil fertilizer application.
Received in revised form 15 January 2018 A set of ten biofortified and non-biofortified wheat varieties developed at the International Maize and Wheat
Accepted 17 January 2018 Improvement Center, Mexico, were used for this study. Two analytical methods were employed to inves-
Available online 31 January 2018
tigate the contents and localization of the trace metals Zn and Fe within the grains, one with polarized
monochromatic X-rays for lower limits of detection, and another featuring polycapillary lenses for micro-
metric beam size (l-EDXRF). Elemental maps were obtained with l-EDXRF allowing for the study of Zn and
Fe localization in plants grown in normal and Zn-enriched soil. It is acknowledged that the biofortification
procedures result in around 30% average increase in overall Zn concentration when compared to other high
Zn genotypes grown in normal soil. A genotypic ranking was performed taking into account the influence
of the measurement methods and field conditions and the obtained results show that two of the top three
varieties regarding zinc contents also rank among the top three in terms of Fe concentration. Elemental
mapping analysis seems to favor the use of integral flour for the manufacture of bread and pasta products,
as the bran retains most of the minerals.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction (XRF) spectrometers are widespread commercial and academic solu-

Biological systems that are complex and heterogeneous have a
large interest for the atomic and molecular spectroscopy community
as they provide a huge challenge in obtaining detailed information of
the inherent microscopic features. This interest has been intensified
towards the elemental analysis of biological tissues whose primary
purpose is to be consumed as food [1–4]. As the benefits and dis-
advantages of having certain nutrients in our diets become clearer,
the need for quantifying and mapping these elements in the bio-
logical systems that host them rises. Bench-top X-Ray Fluorescence
夽 Selected Paper from the Colloquium Spectroscopicum Internationale XL (CSI
2017), Pisa, Italy, 11–16 June 2017.
* Corresponding author.
E-mail address: mguerra@fct.unl.pt (M. Guerra).
tions for elemental analysis [5,6]. They offer fast elemental analysis,
with low costs of operation, no sample preparation or destruction
and a wide number of identifiable elements. Even though compet-
ing techniques such as Inductive Coupled Plasma Mass Spectrom-
etry (ICP-MS) [7], Scanning Electron Microscope coupled to Energy
Dispersive Spectrometry (SEM-EDS) [8], nano Secondary Ion Mass
Spectrometry (nanoSIMS) [9] and Synchrotron Radiation micro-XRF
(SR-XRF) [10], usually offer better lateral resolution and lower limits
of detection they are not as efficient in terms of time, operation cost
and complexity of sample preparation. Cereal grains are heteroge-
neous biologic tissues whose analysis benefits from techniques such
as l-XRF because of the micro sized elemental localization and imag-
ing [11]. Out of all the cereals, rice seems to be the most researched
due to its primordial role in Asian countries’ diet. The in vivo mineral
distribution patterns in rice grains and shifts in those distribution
patterns during progressive stages of germination were analyzed by

https://doi.org/10.1016/j.sab.2018.01.006
0584-8547/© 2018 Elsevier B.V. All rights reserved.
 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79
synchrotron X-ray microfluorescence [7] in parallel with the evalua-
tion of mineral concentrations in whole rice grains, hulls, brown rice,
bran and polished rice by inductively coupled plasma mass spec-
troscopy (ICP-MS). Also, Iwai et al. [12] studied the dynamic changes
in the distribution of some nutritionally important nutrients (P, Ca,
K, Fe, Zn, Cu) in relation to the accumulation of phytic acid during
rice seed development by both inductively coupled plasma optical-
emission spectrometry (ICP-OES) and micro-XRF imaging analysis.
Other elements with well-known effects on public health might also
be detected - As and Hg have been measured and their distribu-
tion mapped in brown rice samples collected in Korea by use of
the ICP-MS with laser ablation [13,14]. In this context, the need
to study the total concentration and the elemental distribution of
particular micronutrients, especially iron and zinc, in other major
food seed crops such as wheat, maize, barley, rye, is a reason of
concern since nutrient dietary deficiency is a worldwide epidemic
that affects more than two billion people according to the World
Health Organization (WHO) [15]. Zinc deficiency is a global nutri-
tional problem, particularly affecting infants, children, adolescents,
pregnant, and lactating women due to the increasing requirements
for the element. Furthermore, zinc deficiency during growth peri-
ods results in growth failure - epidermal, gastrointestinal, central
nervous, immune, skeletal, and reproductive systems are the organs
most affected clinically by zinc deficiency [16]. In this case, deficiency
is linked with a physiological state that increases the Zn require-
ments but in others, deficiencies may well be attributed to both inad-
equate intakes of dietary zinc (rural diets in developing countries are
plant-based mainly) and disease states that induce excessive losses
or impair utilization of Zn [17]. Additionally, clinical diagnosis of
Zn deficiency in humans remains problematic - blood plasma/serum
zinc concentration, dietary intake, and stunting prevalence are the
best known indicators so far [16].
Nutrient supplements are the main medical response in devel-
oped countries but, due to economic reasons, may be of difficult
access in developing countries. They provide high intakes of the
micronutrients needed by the subjects, but may also introduce
unwanted elements to their diet [18]. Biofortified cereals are a
natural solution as staple food constitutes the largest part of the
diet in various countries [2]. Regardless of whether cereals are
biofortified through selective breeding, nutrient rich fertilizers or
genetic manipulation, various studies have found major increments
of selected micronutrients in biofortified species [19–22], although
some authors claim that biofortification with zinc may improve the
serum zinc status of populations if zinc is the sole micronutrient
used for fortification. If zinc is added to food in combination with
other micronutrients, it may make little or no difference to the
serum zinc status [23]. Also, some current agronomic practices where
phosphorus fertilization is used, has a negative effect on the zinc
content of wheat grains due to the antagonism effect of phospho-
rus [24], in a similar way of phytic acid - the storage compound of
phosphorus in seeds, which has the capability of strongly binding
to metallic cations, thus rendering them unavailable for intestinal
mucosa absorption. Hence, the overall Zn content in the grains is
not the only important feature of Zn-biofortification, but also its
potential bioavailability which can also be altered in the food final
production stages [25]. Another important issue of biofortification
studies is that the enrichment of cereal grains or fruits in particular
nutrients for example, must take into account the threshold of tox-
icity since the excess of iron or zinc may reduce yield and interfere
in the ionic balance of other elements [26–28] although soil-plant
interactions also played a crucial role in regulating the mineral load
achieving the desired enriched targets [28,29]. In the current work
we want to show the suitability of bench-top micro-XRF and tri-
axial energy dispersive x-ray fluorescence EDXRF for analysis and
imaging of cereal grains. In particular we will thoroughly study Zn
biofortified wheat grains from different genotypes (10 in total) and
71
compare with non-biofortified control wheat species looking for ele-
ment variation across all species as well as discuss the genotype
ranking of Zn content. Through multi-elemental imaging we seek
to determine where each element is localized and see where Zn in
particular is most prevalent. As we are able to determine not only
the selected nutrients but also other elements, research was also
directed towards understanding how these impact the final product
after food processing. Fe is also a very important nutrient, and in
light of what has been said before about hidden hunger of developing
countries, we have also performed an analysis of Zn/Fe association
as well as genotype ranking for this element. From another stand
point, for food processors it is very important to maximize beneficial
nutrients and eliminate unwanted toxic elements.
2. Materials and methods
2.1. Plants
A set of ten biofortified and non-biofortified wheat varieties devel-
oped at the International Maize and Wheat Improvement Center
(CIMMYT), Mexico were used for this study. Wheat varieties included
in the study are: first semi-dwarf green revolution varieties (Sonalika
and Siete Cerros T66), a synthetic-hexploid derived drought-tolerant
soft wheat variety ‘Vorobey’, and Baj, an early-maturing variety which
in this work will be used as control as it is a non-biofortified but
high-yielding genotype. As another control sample, we will also use
the ‘Borlaug 100’ which is a recently released high-yielding wheat
variety with no previous fortification. In this work, 5 biofortifed high
Zn wheat lines from the biofortification breeding program will be
studied as well, namely Zinc Shakti (Sample #1 and #8) which is a
high Zn and Fe density genotype with +14 lg/g Zn over local vari-
eties and is being grown by more than 250,000 farmers in India, and
Zincol2016 (sample #7) is a recently released high Zn wheat variety
in Pakistan that is being adopted in more than 25,000 ha in Pakistan.
2.2. Field trials
Field trials were conducted at the Norman E. Borlaug Experimen-
tal Station in Ciudad Obregon, Mexico during 2013–14 crop seasons.
The experimental materials (samples 7–11) were produced from
the Zn-enriched area where basal application of 25 kg ha- 1 ZnSO4 •
7H2 O was applied to minimize soil Zn heterogeneity in the exper-
imental fields. The rest of the samples (1–6) was produced under
optimum growing conditions with recommended rates of NPK fer-
tilizers which were applied as basal and top-dressings during the
crop growing season. Optimal water was supplied through five irri-
gation systems during the crop period to favor good plant stand. The
trial was laid out in alpha-lattice design with two checks with two
replicates. Each entry, with approximately 60–70 seeds, was grown
in 1 m long paired rows, spaced 20 cm apart on top of 80 cm wide
raised beds, with a 0.5 m pathway. Pesticides were applied as needed
to keep experimental plots free from diseases and aphids.
2.3. Analytical methods
2.3.1. EDXRF with triaxial geometry
The energy dispersive x-ray fluorescence setup used in this work
consists of a commercial X-ray tube (Philips, PW 1140; 100 kV,
80 mA), equipped with a molybdenum (Mo) changeable secondary
target in a triaxial geometry [30]. With this setup we obtain a nearly
monochromatic source, with the Ka and Kb lines of Mo, with energies
of 17.44 and 19.60 keV. The X-ray tube, the secondary target and the
sample are in a triaxial geometry with 90◦ angles between the tube,
secondary target, sample and the detector in order to decrease the
background by polarization of the incident radiation as well as min-
imizing Rayleigh and Compton scattering [31]. Both the X-ray beam
72 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79
emitted by the secondary target and subsequently by the sample
are collimated throughout two silver tubes in order to reduce the
angular divergence and effectively improving the detection limits.
Due to the out-of-plane incidence, the X-ray spot at the sample
has an elliptic form, with a horizontal axis of approximately 20 mm
and a vertical axis of approximately 15 mm. The X-ray detector is
a liquid nitrogen cooled Si(Li) detector, with a 30 mm2 active area
and 8 lm Be window. The energy resolution is 138 eV for Mn Ka and
the acquisition system is a Nucleus PCA card with pulse processing
and dead time corrections, automatically adjusted by a commercial
pulse processor (Oxford). The operating conditions of this system
for all analyzed spectra were 50 kV and 20 mA. Quantification of
the spectra was performed with an in-house developed code based
on the fundamental parameters method and checked against sev-
eral standard reference materials, such as orchard leaves (NBS-1571),
Poplar Leaves (GBW 07604) and Bush Branches (GBW 07603) which
share a biological matrix with the wheat grains composed essen-
tially of hydrocarbons. The wheat samples were prepared by grinding
6 grains in a mortar and the resulting powder was pressed under
10 tons of force for 2 min. The obtained pellet was then glued with a
hydrocarbon glue onto a mylar sheet and placed on an acrylic frame
which was then positioned at a 90◦ angle between the incoming X-
ray beam and the detector. Three measurements were performed for
each sample in order to minimize statistical errors.
2.3.2. l-EDXRF
All elemental maps of the wheat grains were obtained using
the micro-Energy Dispersive X-Ray Fluorescence (l-EDXRF) system
(M4 TornadoTM , Bruker, Germany). This commercial spectrometer
consists of an air-cooled micro-focus side window Rh-anode X-ray
tube, powered by a low-power HV generator. The system features
poly-capillary X-ray optics, which allow a beam spot size of around
22 lm for Rh Ka . In all of the measurements, the X-ray generator was
operated at 50 kV and 100 lA without the use of filters, in order to
enhance the ionization of low-Z elements without compromising the
peak-to-background ratio for medium-Z elements such as Zn.
Detection of the fluorescence radiation is performed by an
energy-dispersive silicon drift detector, XFlashTM , with a 30 mm2
sensitive area and an energy resolution of 142 eV for Mn Ka . As in
previous studies, the wheat grains were cut in half longitudinally,
along the crease tissue, with a stainless steel, platinum coated razor
blade. The absence of platinum L-lines on the X-ray spectra guar-
antees us that no Fe contamination was induced upon cutting the
grain. The two halves of the longitudinally cut grains were then glued
onto a mylar foil with the inner sides upward, and placed on the
Fig. 1. Elemental map of two grains corresponding to different samples, highlighting the anatomical structure selection procedure. In a) it can be seen that the vascular bundle
was severed upon sample preparation, while in b) one can see that the vascular bundle has remained solely on one side of the cut grain.
M4 Tornado x, y, z translation stage for analysis. Since we were able
to analyze only a single grain with l-EDXRF, due to beamtime con-
straints, and the EDXRF with triaxial geometry required at least 6
ground grains, the measurements across both methods were not per-
formed on the same grain. Measurements were taken under 20 mbar
vacuum conditions and performed directly on the two sides of the
grains. The maps were performed with a pixel spacing of 15 lm, in
order to completely cover the grain, with a measuring time of 6 ms
per pixel, and the final images were created with the built-in soft-
ware from the M4 TornadoTM , ESPRIT. The obtained mass fractions
were not corrected to the total grain’s biomass, and hence the values
for the individual structures within the grain naturally present much
higher concentration than values for the whole grain. This is due to
the fact that the endosperm, which accounts for more than 80% of the
grain biomass, is mainly composed of low atomic weight elements
that cannot be measured by this technique.
The geometrical constraints of the spectrometer do not allow
the X-ray beam to hit the sample perpendicularly, instead the sam-
ple is irradiated from the side at a 50◦ angle, making a 100◦ angle
with the detector. Most of the times, this results in shadows in the
elemental maps, which have to be analyzed bearing this in mind.
With the object area selection tool of the software we were able
to select subsets of the whole elemental map, corresponding to dif-
ferent structures within the grain, namely the embryo, endosperm,
vascular bundle as well as the whole grain as can be seen in Fig. 1. In
the cases where either the vascular bundle or the embryo was sev-
ered upon slicing of the grain, the structures, easily identified in the
elemental maps, were properly delimited within the maps in order
to obtain a more reliable quantification. In Fig. 1 a) it is easily seen
that the vascular bundle was sliced during sample preparation, and
the chosen areas for quantification, on both halves of the grain, were
chosen so that only the signal from that structure was processed.
Quantification of the pixel sum spectra of these selected regions was
performed with the fundamental parameters method of the ESPRIT
software.
3. Results and discussion
3.1. Whole grain
For the whole grain quantification, we employed both the l-
EDXRF method and the high-sensitivity analytical method EDXRF
with a triaxial geometry as described in the Materials and Methods
Section. Both methods will be labeled M1 and M2 from now on for
simplicity. The final mass fractions for the whole grain were then
 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79 73

Table 1 be related, not only to the grain to grain variation within a spike [32]
Concentration group means and standard errors of Zn and Fe for genotypes (across for Zn and Fe (note that the measurements of both methods were not
methods and soils), methods (across genotypes and soils) and soils (across genotypes
and methods) in the case of the whole grain data. The notation nG , nM , nN and nE
performed on the same grain), but also to the fact that the sample
refer to the number of measurements used to compute the means in the case of the preparation for the elemental mapping in M1 exposes the vascu-
genotypes, methods, normal and enriched soils, respectively. lar part and embryo of the grain to the excitation beam, effectively
Genotype Zn (lg/g) Fe (lg/g) increasing the metal concentration readings. As Zn is much more
spread out throughout the grain than Fe, and it typically has a slightly
G1 82(17) 93(30)
lower concentration, the lack of self-absorption within the exposed
G2 73(8) 61(24)
G3 77(1) 48(12) layers leads to an apparent enhancement of the Fe concentrations.
G4 65(10) 54(10) Since we cannot rule out some other effect being at play here, the
G5 63(3) 46(12) genotype concentrations have been calculated as a weighted aver-
G6 57(9) 43(14) age of the M1 and M2 results, taking into account the sample size
G7 40(8) 68(31)
G8 80(15) 80(37)
for each individual measurement (as explained before). Poor cali-
G9 70(10) 65(34) bration within the quantification procedures has been ruled out by
G10 68(23) 64(30) comparing quantification results of the 3 standard reference mate-
nG = 2 concentrations per genotype (G2,. . . ,G10); rials referred to in Section “EDXRF with triaxial geometry” with the
nG1 = 4 concentrations for genotype G1.
two methods. Note, however, that the standard reference materi-
als are sold in powder form and a pellet was made for analysis.
Method Zn (lg/g) Fe (lg/g)
This means that the assumptions of the quantification methodolo-
M1 74(17) 81(23) gies were met, for both techniques, regarding the standard reference
M2 64(13) 49(20) materials, unlike what happens with the analysis of a sliced grain
nM = 11 concentrations per method.
with method M1. Here, the greatest differences are observed in the
cases of genotypes G6-G10, which are the ones coming from the zinc
Soil Zn (lg/g) Fe (lg/g)
enriched soil. In addition, measurements obtained from M1 show
Normal 63(14) 60(25) more variability (Table 1; Method - standard errors), which should
Enriched 76(16) 71(28) be related to the fact that only one grain was analyzed with M1 for
nN = 6 ∗ 2 & nE = 5 ∗ 2 concentrations per field. each genotype.
The presented error bars in Figs. 3 and 4 represent the quadrature
Genotype coding
sum of the instrumental uncertainty and the standard deviation of all
G1 Zinc-shakti measurements. From Fig. 3 it is clear that both in the non-biofortified
G2 Sonalika plants as well as in the Zn-rich soil grown plants, the majority of Zn
G3 Siete Cerros
is deposited within the vascular bundle and embryo (see also Fig. 5).
G4 Vorobey
G5 Baj Comparison of overall Zn concentration in the whole grains (Fig. 4a)
G6 Borlaug 100 showed 10–35 % higher values over historical varieties of Siete Cer-
G7 Zincol 2016 ros T66, Vorobey and Sonalika over the early maturing control Baj,
G8 Danphe whereas 55 to 90% higher Zn when compared to recently released
G9 Mayil
high yielding variety Borlaug 100. About 28% higher Zn in the biofor-
G10 Hgo
tified Zinc-shakti grown under normal soil Zn levels (sample 1) – the
Method coding
first high Zn variety released in India, over the check variety Baj while
for Zinc-shakti grown under Zn-enriched soil (sample 8) there was an
M1 l-EDXRF approximately 60% increase in Zn, followed by 40% higher Zn in Zin-
M2 triaxial-EDXRF
col 2016, a high Zn wheat variety released in Pakistan. Similarly, an
81%, 98% and 125% higher Zn of samples 1, 7 and 8, respectively, over
obtained through a weighted average of the M1 and M2 results to the control Borlaug100 was observed. The other 3 biofortified lines
take into account the sample size for each technique. (Danphe, Mayil and Hgo) showed 70–82 % higher Zn over Borlaug100
The results produced by method M1 show, in general, slightly and 19–27 % increase over the Baj genotype. The synthetic-hexploid
higher readings of both Zn and Fe concentrations than the ones derived drought-tolerant soft wheat variety Vorobey, showed 10 and
obtained via method M2 (Table 1; Fig. 2), although for Zn the quite 56% higher Zn over Baj and Borlaug100, respectively. The first semi-
sizable error bars overlap for several genotypes. This disparity may dwarf green revolution wheat variety Sonalika showed 34% and 90%

Fig. 2. Concentrations of iron (Fe) and zinc (Zn) in each genotype as measured by methods M1 and M2. The measurements with M1 were made on a single grain, while M2 shows
one measurement of a homogeneous mixture of 6 grains. The error bars refer only to the experimental uncertainties of the quantification procedure as they represent only one
measurement.
74 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79

Fig. 3. Average Zn concentration, in lg/g, of all of the studied samples, divided by anatomical structure. The error bars represent the quadrature sum of the instrumental
uncertainties with one standard deviation.

higher Zn over Baj and Borlaug100, even when grown in a non-Zn- within the whole grain, when compared to the same genotype grown
rich environment, outperforming the results of the Zinc-shakti which in normal conditions without significant loss of crop yield. Unlike
featured a 27% and 81% increase in Zn. The biofortification of the other genotypes such as the Triticum aestivum cv. Roxo where a 10-
Zinc-shakti genotype showed a 25% increase in Zn concentration, fold increase in grain Zn content, from the zeroth non-biofortified

Fig. 4. Zn concentration, in lg/g, of all of the studied samples, divided by anatomical structure. The average values of the samples grown in the Zn-rich soil as well as in normal
soil are shown as blue and red solid lines, respectively. The dashed lines represent one standard deviation.
 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79 75

Fig. 5. Zn and Fe concentrations by kernel anatomic structure in lg/g: EM - embryo; EN - endosperm; V - vascular bundle. G1 values are the mean values from both normal and
enriched soils.

generation to the second generation (under growth chamber condi-
tions) was achieved [20], the increase in the Zinc-shakti genotype
due to biofortification is modest. This can be explained by the fact
that for Zn, the bioaccumulation in the grain is limited by transport
barriers from the rachis to the vascular bundle, such as the xylem
discontinuity in the maternal tissue [33,34], which reduces the effi-
ciency of soil-applied Zn biofortification, as well as the fact that this
genotype is already a result of several biofortified generations, and its
metabolic elasticity might be reaching a saturation point. This effect
was also seen for Triticum aestivum cv. Roxo in reference [20] for
higher generations of grains undergoing consecutive Zn biofortifica-
tion. The high Zn biofortified wheat lines present significantly higher
Zn concentrations over the normal varieties (Baj and Borlaug 100)
and the historical varieties also feature higher Zn concentrations
even when grown in normal soil with normal fertilization.

Table 2
Concentration group means and standard errors of Zn and Fe for genotypes (across 3.2. Grain structures
structures and soils), anatomical structures (across genotypes and soils) and soils
(across genotypes and structures) in the case of the structural grain data. The notation As explained in the Materials and Methods Section, three struc-
nG , nL , nN and nE refer to the number of measurements used to compute the means in tures within the grain halves were pinpointed by selecting subsets of
the case of the genotypes, structures, normal and enriched soils, respectively.
the pixel spectra from the elemental maps as shown in Fig. 1.
Genotype Zn (lg/g) Fe (lg/g) From Fig. 4b) one observes a significant increase in Zn deposi-
G1 180(150) 170(140) tion within the vascular bundle of the analyzed grains that were
G2 139(95) 106(92) subjected to the biofortification procedure when compared to those
G3 78(54) 110(130) grown in non-Zn-rich soils. As this organ is the primary transport
G4 133(93) 120(140)
vehicle of nutrients to the grain, this result is not surprising, how-
G5 107(71) 90(110)
G6 82(48) 141(74) ever the deposition of Zn within other structures of the grain does
G7 170(130) 240(180) not show a very strong correlation with these values (still it is higher
G8 140(110) 170(160) in the embryo than in the endosperm). This is indicative that the
G9 140(95) 170(200) transport to the grain is very difficult in soil-biofortified plants, nev-
G10 104(80) 120(140)
nG = 3 concentrations per genotype (G2,. . . ,G10);
ertheless, the direct connection of the vascular bundle to the embryo,
nG1 = 6 concentrations for genotype G1. seems to favor Zn uptake when compared to the endosperm for
which there is no clear trend. In foliar Zn application, however, espe-
Structure Zn (lg/g) Fe (lg/g) cially if applied at early milk/early dough, the metal transport to
the endosperm is much higher than soil fertilized biofortification as
Embryo 189(57) 282(64)
reported by Ajiboye et al. [35] and complemented by our results. The
Endosperm 27.3(5.2) 34(16)
V. bundle 180(83) 130(100) embryo, which is directly connected to the rachila, has a preferen-
nL = 11 concentrations per anatomical structure. tial uptake of rachis-transported nutrients, similarly to the vascular
bundle, which can be directly concluded from the results presented
Soil Zn (lg/g) Fe (lg/g) in Figs. 3 and 4 (bottom-left and top-right corners). Similarly to
the vascular bundle, the biofortification procedure employed in this
Normal 110(70) 124(98)
Enriched 160(120) 180(150)
nN = 6 ∗ 3 & nE = 5 ∗ 3 concentrations per soil. Table 3
Results from the classical and robust fits of model (1).
Genotype coding Type of fit Estimated variances Statistical tests
G1 Zinc-shakti
p-values
G2 Sonalika
G3 Siete Cerros ss2 2
sm se2 Shapiro-Wilks test Levene test
G4 Vorobey
G5 Baj Fe Classical 96.4 466.1 222.6 0.5652 0.0032
G6 Borlaug 100 Robust 0.00 658.9 250.6 − −
G7 Zincol 2016 Zn Classical 109.4 48.0 98.2 0.7386 0.0573
G8 Danphe Robust 159.3 69.4 101.8 − −
G9 Mayil
Note: Statistical tests report p-values for the normality and homoscedasticity of the
G10 Hgo
errors tests, respectively, Shapiro-Wilks and Levene’s tests.
76 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79
work shows a slight average increase of the total Zn concentration in
the embryo specially when compared to the control grains Baj and
Borlaug100. Curiously, the overall high Zn mass fraction exhibited
by the control Zinc-shakti genotype results from a high Zn content
embryo, as the other structures (vascular bundle included) do not
show very high Zn concentrations, similar to the non-biofortified
controls. Nevertheless, the Zn concentration within the embryo,
still increases by 40% upon biofortification for this particular geno-
type. Regarding the endosperm, which accounts for around 80% of
the total biomass of the grain, and corresponds to the grain part
which is transformed in the white flour for human consumption,
the results show that there is no significant change between bio-
fortified and non-biofortified species, independently of the chosen
genotype. In fact, for the Zinc-shakti genotype there seems to be a
34% increase in the Zn concentration within the endosperm, but the
error bars slightly overlap. Still, as the endosperm accounts for most
of the grain’s biomass, even a slight increase in concentration results
in a total Zn mass that might be substantial. If the bioavailability
of Zn located in the endosperm is high, it can be even better as a
way for improving Zn dietary intake problems for high risk
populations.
Fig. 5 shows how Zn and Fe concentrations deposited in each of
the three grain structures and Table 2 further complements infor-
mation regarding these concentrations by summarizing the group
means and standard deviations of the concentration for genotypes,
soils and structures.
3.3. Statistical analysis
In this section we study the relationship between Zn and Fe
concentrations at the whole grain level as well as at the level of
embryo, endosperm and vascular bundle structures. A ranking of all
the studied genotypes with regard to Zn and Fe concentrations is
also performed. One should note that the statistical results reported
below are merely indicative, as the available sample size is small, a
fact that was related to both laboratory and beam time limitations
and that could not be overcome.
Herein, varieties/genotypes Zinc-Shakti, Sonalika, Siete Cerros T66,
Vorobey, Baj, Borlaug100, Zincol 2016, Danphe, Mayil and Hgo are
coded as G1,. . . ,G10, (see Table 1). Note that variety G1 is the only
variety grown both in the non-Zn-rich as well as the Zn-rich soil
(normal and enriched soils).
3.4. Whole grain analysis
Because, in this section, interest lies in studying the linear relation-
ship between Fe and Zn concentrations, we begin by fitting the linear
Table 4
Genotype estimates and ranking after the classical and robust fits of model (1).
Concentration values are in lg/g.
∗
Genotypes G1–G10 sorted in ascending order; gray cells refer to ranking
disagreement;
Note: The residuals from the classical fits do not violate the normality assumption.
mixed effects model (LMM; Eq. (1)) to the Fe and Zn data, separately,
in order to compute the genotypic mean values of Fe and Zn.
y = Genotype + (1 | Soil) + (1 | Method) + e, e ∼ N(0, s 2 I). (1)
iid
This procedure allows for the removal of the effects, should there
be any, of soil and method from the Fe and Zn concentration varia-
tion when computing the genotypic means. Because the classical fit
of the LMM may produce biases in the estimation when one or more
of its underlying assumptions are violated (e.g., independent, nor-
mal and homoscedastic errors), a robust approach is also considered
in this study. In general, robust methods produce similar results to
the classical approach when data conforms to model’s premises, and
allow for more stable results when the data deviate from those. A
nice review on these methods can be found in [36].
Results regarding variance components estimates and model val-
idation are displayed in Table 3. Both Fe and Zn fitted models pass
the normality test of Shapiro-Wilks (p − value  0.05). However,
in the case of fit of Fe, there is evidence against the homoscedastic-
ity assumption of the errors, with the Levene’s test rejecting the null
hypothesis (p − value  0.05). This fact might be enough to justify
the differences in results between classical and robust approaches.
Here, the robust fit shows that, contrary to what is stated by the clas-
sical fit, soil has no part at all in explaining the variation in the Fe
concentrations (robust ss2 = 0). Nevertheless, both results agree that
method is well included in the model as a random effect (sm 2
≥ se2 ).
In the case of the fit of Zn, Levene’s null hypothesis is accepted at the
5% significance level by a hair’s breadth (p − value > 0.05). In this
case, the classical and robust results agree that soil plays a major role
in the explanation of the Zn concentrations variation (ss2 ≥ se2 ), an
outcome that was expected since soils differ in the sense that one is
normal and the other one is Zn enriched. Approaches also agree that
method has little part in the explanation of this variation (sm 2
≤ se2 ).
After the Fe and Zn genotypic means are estimated (fixed effects in
model (1)), the genotypes are ranked from low to high concentration
levels. Results are displayed in Table 4. We acknowledge that classical
and robust results disagree, in terms of the ranking of the genotypes,
only in the case of the Fe concentrations (gray cells). Still, dissent is
only seen in the case of genotype G6, which refers to the recently
high-yield released control wheat variety Borlaug 100, and could be
due to the violation of the error homoscedastic premise reported
earlier on. Additionally, the range of robust Fe genotypic means is
smaller than the classical. In the case of Zn concentrations, there is
complete agreement between both classical and robust approaches
and the range of Zn genotypic values is also practically the same. Here,
in contrast with what happens with Fe concentrations, genotype G6
ranks last, which is an indication of the specific low-Zn content of this
particular variety.
In the case of Fe, with the exception of genotype G6, which refers
to a variety that has never been fortified before, higher levels of
concentrations are observed in the genotypes which were assigned
to the enriched zinc soil (G7–G10; G1 was assigned to both soils).
Table 5
Results from the classical and robust fits of model (2).
Type of fit Estimated variances Statistical tests
p-values
2
ss2 sstr se2 Shapiro-Wilks test Levene test
Fe Classical 0.00 15,287 4084 0.1389 0.9072
Robust 0.00 21,833 3089 − −
Zn Classical 6253 8071 2147 0.4546 0.6874
Robust 8490 10,571 2147 − −
Note: Statistical tests report p-values for the normality and homoscedasticity of the
errors tests, respectively, Shapiro-Wilks and Levene’s tests.
 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79 77

Table 6
Genotype estimates and ranking after the classical and robust fits of model (2). Concentration values are in
lg/g.

∗
Genotypes G1–G10 sorted in ascending order; gray cells refer to ranking disagreement;
Note: The residuals from the classical fits do not violate the normality assumption.

Notwithstanding, data information is insufficient to conclude that
soil has an effect on the levels of Fe concentration. This could be hap-
pening due to chance alone or be genotype specific and therefore
further study is needed. In particular, the same genotypes should be
replicated across the two soils, which was not the case in this study.
To analyze the relationship between Fe and Zn, the Pearson corre-
lation test, Kendall, Spearman [37,38] and another robust correlation
tests [39] were computed. All tests agree in favor of the null hypoth-
esis, i.e., that true correlation is equal to zero. Indeed, three of the
methods compute a correlation 0.11, which is usually considered
as “very weak” correlation.
3.5. Grain structures analysis
From both the visual output (Fig. 5) and the summary statistics
(Table 2), we acknowledge that Fe and Zn tend to deposit less in
the endosperm, which was to be expected [11,35]. Also, while Fe
deposits in higher quantities (≥200 lg/g) in the embryo, no clear
trend is seen for the higher concentrations of Zn (≥200 lg/g), which
deposited both in the embryo and vascular bundle structures. Addi-
tionally, it can be seen how much the concentration of Zn increases
for genotype G1 from the normal to the Zn enriched soil. Quite curi-
ously, at the vascular bundle structure level, the concentration of Fe
for this genotype decreases from the normal to the Zn enriched soil
thus contradicting the observed general trend.
The computation of the Fe and Zn genotypic means via the robust
Zn fitted models pass the normality test of Shapiro-Wilks and the
Levene’s test of the homoscedasticity of the errors. Results between
approaches are also similar. In the case of Fe only structure has signif-
2
icant effect in the random effects component (sstr ≥ se2 ). In the case
of Zn both structure and soil have significant effects in the random
2
effects component (ss2 , sstr ≥ se2 ) as expected, since both the soil fer-
tilization and the physiology of the grains can significantly influence
the Zn uptake.
Table 6 shows that the rankings for Zn, again, agree between clas-
sical and robust approaches. In the case of Fe, disagreement is only
seen in the ranking of genotypes G1 and G8, which interchange. This
only occurs because the computed Fe genotypic mean for G1 differs
in −2.1 units from the classical to the robust results, which, at this
scale can be considered to be negligible.
As to the correlation between Fe and Zn, the Pearson correlation
test, Kendall, Spearman [37,38] and the robust correlation test of [39]
agree in favor of the null hypothesis, i.e., that true correlation is, sim-
ilarly to the whole grain results, equal to zero. However, the three
methods now compute a negative correlation of −0.30, which is
usually considered as “weak” correlation. Fig. 6 shows how poorly
the classical and robust fits of the linear model (3), describe the data,
illustrating the lack of correlation.
Fe = Zn + e, e ∼ N(0, s 2 I). (3)
iid

and classical fits of model (2) leads to the genotype rankings reported
in Table 6.
y = Genotype + (1 | Soil) + (1 | Structure) + e, e ∼ N(0, s 2 I).
iid
Results regarding variance components estimates and model val-
idation are displayed in Table 5. It is acknowledged that both Fe and
We further observe that the ranking of genotypes obtained in
Tables 2 and 6 is not exactly the same. Nevertheless, for Fe, the top 3
genotypes overlap in G1, G7 (robust results) and in the case of Zn, the
(2)
top 3 genotypes overlap in G1, G2 (both classical and robust results).
It should also be underlined that in the case of the structural grain
analysis, not all the concentration of Fe and Zn was indeed measured
since a part of the grain was discarded (such as the bran and part of

Fig. 6. Plot of Zn vs Fe concentrations together with fitted classical (red) and robust (dashed blue) regression lines, plus the 95% confidence region (gray) computed from the
classical fit.
78 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79
the endosperm lying below the exposed bran). Because of this lack of
information, to what adds the fact that only one grain per genotype
was analyzed with method M1 and the observed higher standard
errors, we conclude that the whole grain genotypic ranking results
seem more plausible. Although further study is needed in order to
fully assess the varieties with better ability to retain Zn and Fe, these
results already provide helpful and important information, hinting
so far to the existence of varieties that are able to retain high concen-
trations of Zn and Fe simultaneously, and secondly, suggesting that a
compromise between Zn and Fe concentrations needs to be achieved
as no significant correlation between those was observed.
4. Conclusions
In this work we have analyzed normal and biofortified wheat
varieties grown under natural soil and zinc-enriched soil through
fertilization. Several popular wheat varieties, as well as some histor-
ical genotypes, were included in the field trials. It is acknowledged
that the biofortification procedures result in around a 30% aver-
age increase in overall Zn concentration when compared to other
high Zn genotypes, and much higher (35–90 %) when compared
to the control grains Baj and Borlaug100. It is also shown that
this increase in average concentration arises from a big increase in
Zn content within the embryo and vascular bundle, with no sig-
nificant changes, on average, in the endosperm. Nevertheless, for
Zinc-shakti, which is a very popular variety in developing countries,
we found a 34% increase in concentration within the endosperm,
which is a substantial change specially taking into account that the
endosperm accounts for almost 80% of the grain’s total biomass. The
high nutrient contents on the Vascular bundle and embryo, both in
the biofortified lines and those which were regularly grown, seem to
favor the use of integral wheat flour particularly in countries where
low Zn diets are a major public health concern. The statistical analy-
sis further hints that varieties Zinc-shakti, Sonalika, and Zincol2016
are the top 3 varieties regarding zinc contents with Zinc-shakti, and
Zincol2016 also ranking in the top 3 in terms of Fe concentration.
Contrary to the classical results, the robust rankings of the genotypes
in terms of Fe concentrations, also point to Borlaug100 as a top 3
variety, contrasting with its specific Zn ranking, which comes last.
Acknowledgments
P. Cardoso and M. Guerra acknowledge the support of the
Portuguese Fundação para a Ciência e Teconologia, FCT, under
contracts nos. PDE/BDE/114483/2016 and SFRH/BPD/92455/2013,
respectively. This work was supported in part by the research
center grant no. UID/FIS/04559/2013 to LIBPhys-UNL, from the
FCT/MCTES/PIDDAC. We greatly acknowledge the additional funding
support from HarvestPlus challenge program-OPP1019962 (partly
funded by Bill and Melinda Gates Foundation and others) for the
research work at CIMMYT, Mexico.
References
[1] E. Lombi, E. Smith, T.H. Hansen, D. Paterson, M.D. de Jonge, D.L. Howard,
D.P. Persson, S. Husted, C. Ryan, J.K. Schjoerring, Megapixel imaging of
(micro)nutrients in mature barley grains, J. Exp. Bot. 62 (1) (2011) 273–282.
[2] K.R. Trijatmiko, C. Dueñas, N. Tsakirpaloglou, L. Torrizo, F.M. Arines, C. Adeva,
J. Balindong, N. Oliva, M.V. Sapasap, J. Borrero, J. Rey, P. Francisco, A. Nelson,
H. Nakanishi, E. Lombi, E. Tako, R.P. Glahn, J. Stangoulis, P. Chadha-Mohanty,
A.A.T. Johnson, J. Tohme, G. Barry, I.H. Slamet-Loedin, Biofortified indica rice
attains iron and zinc nutrition dietary targets in the field, Sci. Rep. 6 (2016)
19792.
[3] H. Gallardo, I. Queralt, J. Tapias, M. Guerra, M.L. Carvalho, E. Marguí,
Possibilities of low-power X-ray fluorescence spectrometry methods for rapid
multielemental analysis and imaging of vegetal foodstuffs, J. Food Compos.
Anal. 50 (2016) 1–9.
[4] I. Cakmak, M. Kalayci, Y. Kaya, A.A. Torun, N. Aydin, Y. Wang, Z. Arisoy,
H. Erdem, A. Yazici, O. Gokmen, L. Ozturk, W.J. Horst, Biofortification
and localization of zinc in wheat grain, J. Agric. Food Chem. 58 (16) (2010)
9092–9102.
[5] M. Guerra, C. Ferreira, M.L. Carvalho, J.P. Santos, S. Pessanha, Distribution of
toxic elements in teeth treated with amalgam using l-energy dispersive X-ray
fluorescence, Spectrochim. Acta B At. Spectrosc. 122 (2016) 114–117.
[6] M. Guerra, S. Longelin, S. Pessanha, M. Manso, M.L. Carvalho, Development
of a combined portable x-ray fluorescence and Raman spectrometer for in situ
analysis, Rev. Sci. Instrum. 85 (6). (2014)
[7] L. Lu, S. Tian, H. Liao, J. Zhang, X. Yang, J.M. Labavitch, W. Chen, Analysis of
metal element distributions in rice (Oryza sativa L.) seeds and relocation during
germination based on X-ray fluorescence imaging of Zn, Fe, K, Ca, and Mn, PLoS
ONE 8 (2) (2013) e57360.
[8] R.C. Power, D.C. Salazar-García, R.M. Wittig, A.G. Henry, Assessing use and
suitability of scanning electron microscopy in the analysis of micro remains in
dental calculus, J. Archaeol. Sci. 49 (2014) 160–169.
[9] K.L. Moore, M. Schröder, E. Lombi, F.-J. Zhao, S.P. McGrath, M.J. Hawkesford,
P.R. Shewry, C.R.M. Grovenor, NanoSIMS analysis of arsenic and selenium in
cereal grain, New Phytol. 185 (2) (2010) 434–445.
[10] B. Kyriacou, K.L. Moore, D. Paterson, M.D. de Jonge, D.L. Howard, J. Stangoulis,
M. Tester, E. Lombi, A.A.T. Johnson, Localization of iron in rice grain using
synchrotron X-ray fluorescence microscopy and high resolution secondary ion
mass spectrometry, J. Cereal Sci. 59 (2) (2014) 173–180.
[11] I. Ramos, I.M. Pataco, M.P. Mourinho, F. Lidon, F. Reboredo, M.F. Pessoa,
M.L. Carvalho, J.P. Santos, M. Guerra, Elemental mapping of biofortified wheat
grains using micro X-ray fluorescence, Spectrochim. Acta B At. Spectrosc. 120
(2016) 30–36.
[12] T. Iwai, M. Takahashi, K. Oda, Y. Terada, K.T. Yoshida, Dynamic changes in
the distribution of minerals in relation to phytic acid accumulation during rice
seed development, Plant Physiol. 160 (4) (2012) 2007–2014.
[13] S.H. Choi, J.S. Kim, J.Y. Lee, J.S. Jeon, J.W. Kim, R.E. Russo, J. Gonzalez, J.H. Yoo,
K.S. Kim, J.S. Yang, K.S. Park, Analysis of arsenic in rice grains using ICP-MS
and fs LA-ICP-MS, J. Anal. At. Spectrom. 29 (7) (2014) 1233–1237.
[14] B. Meng, X. Feng, G. Qiu, C.W.N. Anderson, J. Wang, L. Zhao, Localization
and speciation of mercury in brown rice with implications for Pan-Asian public
health, Environ. Sci. Technol. 48 (14) (2014) 7974–7981.
[15] WHO, The world health report. Reducing risks, promoting healthy life, Tech.
rep., World Health Organization. 2002,
[16] N. Roohani, R. Hurrell, R. Kelishadi, R. Schulin, Zinc and its importance for
human health: an integrative review, Journal of Research in Medical Sciences:
The Official Journal of Isfahan University of Medical Sciences 18 (2) (2013)
144–157.
[17] R.S. Gibson, Zinc: the missing link in combating micronutrient malnutrition in
developing countries, Proc. Nutr. Soc. 65 (1) (2007) 51–60.
[18] P. Cardoso, P. Amaro, J.P. Santos, J.T. de Assis, M.L. Carvalho, Determination of
nickel and manganese contaminants in pharmaceutical iron supplements using
energy dispersive X-ray fluorescence, Appl. Spectrosc. 71 (3) (2016) 432–437.
[19] G. Velu, I. Ortiz-Monasterio, I. Cakmak, Y. Hao, R.P. Singh, Biofortification
strategies to increase grain zinc and iron concentrations in wheat, J. Cereal Sci.
59 (3) (2014) 365–372.
[20] F.C. Lidon, A.S. Almeida, A.R. Costa, A.S. Bagulho, P. Scotti-Campos, J.N.
Semedo, B. Maçãs, J. Coutinho, N. Pinheiro, C. Gomes, A.E. Leitão, I.P. Pais,
M.M. Silva, F.H. Reboredo, M.F. Pessoa, J.C. Ramalho, Sequential zinc and
iron biofortification of bread-wheat grains: from controlled to uncontrolled
environments, Crop Pasture Sci. 66 (2015) 1097–1104.
[21] I.M. Pataco, F.C. Lidon, I. Ramos, K. Oliveira, M. Guerra, M.F. Pessoa, M.L.
Carvalho, J.C. Ramalho, A.E. Leitão, J.P. Santos, P.S. Campos, M.M. Silva,
I.P. Pais, F.H. Reboredo, Biofortification of durum wheat (Triticum turgidum
L. ssp. durum (Desf.) Husnot) grains with nutrients, J. Plant Interact. 12 (1)
(2017) 39–50.
[22] M.W. Blair, Mineral biofortification strategies for food staples: the example of
common bean, J. Agric. Food Chem. 61 (35) (2013) 8287–8294.
[23] D. Shah, H.S. Sachdev, T. Gera, L.M. De-Regil, J.P. Pena-Rosas, Fortification of
staple foods with zinc for improving zinc status and other health outcomes in
the general population, Cochrane Database Syst. Rev. 9 (6) (2016) Cd010697.
[24] Y.-Q. Zhang, Y. Deng, R.-Y. Chen, Z.-L. Cui, X.-P. Chen, R. Yost, F.-S. Zhang,
C.-Q. Zou, The reduction in zinc concentration of wheat grain upon increased
phosphorus-fertilization and its mitigation by foliar zinc application, Plant Soil
361 (1) (2012) 143–152.
[25] V. Ciccolini, E. Pellegrino, A. Coccina, A.I. Fiaschi, D. Cerretani, C. Sgherri, M.F.
Quartacci, L. Ercoli, Biofortification with iron and zinc improves nutritional and
nutraceutical properties of common wheat flour and bread, J. Agric. Food Chem.
65 (27) (2017) 5443–5452.
[26] F. Reboredo, Cu and Zn uptake by Halimione portulacoides (L.) Aellen. A
long-term accumulation experiment, Bull. Environ. Contam. Toxicol. 46 (3)
(1991) 442–449.
[27] F. Reboredo, Interaction between copper and zinc and their uptake by
Halimione portulacoides (L.) Aellen, Bull. Environ. Contam. Toxicol. 52 (4) (1994)
598–605.
[28] F. Reboredo, Zinc compartmentation in Halimione portulacoides (L.) Aellen and
some effects on leaf ultrastructure, Environ. Sci. Pollut. Res. 19 (7) (2012)
2644–2657.
[29] F.H.S. Reboredo, C.A.G. Ribeiro, Vertical distribution of Al, Cu, Fe and Zn in the
soil salt marshes of the Sado estuary, Portugal, Int. J. Environ. Stud. 23 (3-4)
(1984) 249–253.
 P. Cardoso et al. / Spectrochimica Acta Part B 141 (2018) 70–79
[30] S. Pessanha, M. Alves, J.M. Sampaio, J.P. Santos, M.L. Carvalho, M. Guerra, A
novel portable energy dispersive X-ray fluorescence spectrometer with triaxial
geometry, J. Instrum. 12 (01) (2017) P01014.
[31] M. Guerra, M. Manso, S. Pessanha, S. Longelin, M.L. Carvalho, Theoretical and
experimental study on the angular dependence of scattering processes in X-ray
fluorescence systems, X-Ray Spectrom. 42 (5) (2013) 402–407.
[32] D.F. Calderini, I. Ortiz-Monasterio, Grain position affects grain macronutrient
and micronutrient concentrations in wheat, Crop Sci. 43 (1) (2003) 141–151.
[33] T.P. O’Brien, M.E. Sammut, J.W. Lee, M.G. Smart, The vascular system of the
wheat spikelet, Funct. Plant Biol. 12 (5) (1985) 487–511.
[34] S.Y. Zee, T.P. O’Brien, A special type of tracheary element associated with
“Xylem Discontinuity” in the floral axis of wheat, Aust. J. Biol. Sci. 23 (4) (1970)
783–792.
79
[35] B. Ajiboye, I. Cakmak, D. Paterson, M. de Jonge, D. Howard, S. Stacey, A. Torun,
N. Aydin, M. McLaughlin, X-ray fluorescence microscopy of zinc localization
in wheat grains biofortified through foliar zinc applications at different growth
stages under field conditions, Plant Soil 392 (1-2) (2015) 357–370.
[36] R.A. Maronna, R.D. Martin, V.J. Yohai, Robust statistics, theory and methods,
Robust Statistics, John Wiley & Sons, Ltd. 2006, pp. i–xx.
[37] M. Hollander, D.A. Wolfe, Nonparametric Statistical Methods, John Wiley &
Sons, New York, 1973.
[38] D.J. Best, D.E. Roberts, Algorithm AS 89: the upper tail probabilities of
Spearman’s Rho, J. R. Stat. Soc. Ser. C. Appl. Stat. 24 (3) (1975) 377–379.
[39] U. Bodenhofer, F. Klawonn, Robust rank correlation coefficients on the basis of
fuzzy, Mathware Soft Comput. 15 (1) (2008) 5–20.