ArticlePDF Available

Abstract and Figures

The analysis of pollen chemical composition is important to many fields, including agriculture, plant physiology, ecology, allergology, and climate studies. Here, the potential of a combination of different spectroscopic and spectrometric methods regarding the characterization of small biochemical differences between pollen samples was evaluated using multivariate statistical approaches. Pollen samples, collected from three populations of the grass Poa alpina, were analyzed using Fourier-transform infrared (FTIR) spectroscopy, Raman spectroscopy, surface enhanced Raman scattering (SERS), and matrix assisted laser desorption/ionization mass spectrometry (MALDI-TOF MS). The variation in the sample set can be described in a hierarchical framework comprising three populations of the same grass species and four different growth conditions of the parent plants for each of the populations. Therefore, the data set can work here as a model system to evaluate the classification and characterization ability of the different spectroscopic and spectrometric methods. ANOVA Simultaneous Component Analysis (ASCA) was applied to achieve a separation of different sources of variance in the complex sample set. Since the chosen methods and sample preparations probe different parts and/or molecular constituents of the pollen grains, complementary information about the chemical composition of the pollen can be obtained. By using consensus principal component analysis (CPCA), data from the different methods are linked together. This enables an investigation of the underlying global information, since complementary chemical data are combined. The molecular information from four spectroscopies was combined with phenotypical information gathered from the parent plants, thereby helping to potentially link pollen chemistry to other biotic and abiotic parameters.
Content may be subject to copyright.
Combining Chemical Information
From Grass Pollen in Multimodal
Characterization
Sabrina Diehn
1,2
, Boris Zimmermann
3
, Valeria Tantseva
3
, Stephan Seifert
1,2
,
Murat Bag˘cıog˘lu
3
, Mikael Ohlson
4
, Steffen Weidner
2
, Siri Fjellheim
5
, Achim Kohler
3,6
and Janina Kneipp
1,2
*
1
Department of Chemistry, Humboldt-Universität zu Berlin, Berlin, Germany,
2
BAM Federal Institute for Materials Research
and Testing, Berlin, Germany,
3
Faculty of Science and Technology, Norwegian University of Life Sciences, Ås, Norway,
4
Faculty of Environmental Sciences and Natural Resource Management, Norwegian University of Life Sciences, Ås, Norway,
5
Faculty of Biosciences, Norwegian University of Life Sciences, Ås, Norway,
6
Noma AS, Ås, Norway
The analysis of pollen chemical composition is important to many elds, including
agriculture, plant physiology, ecology, allergology, and climate studies. Here, the
potential of a combination of different spectroscopic and spectrometric methods
regarding the characterization of small biochemical differences between pollen samples
was evaluated using multivariate statistical approaches. Pollen samples, collected from
three populations of the grass Poa alpina, were analyzed using Fourier-transform infrared
(FTIR) spectroscopy, Raman spectroscopy, surface enhanced Raman scattering (SERS),
and matrix assisted laser desorption/ionization mass spectrometry (MALDI-TOF MS). The
variation in the sample set can be described in a hierarchical framework comprising three
populations of the same grass species and four different growth conditions of the parent
plants for each of the populations. Therefore, the data set can work here as a model
system to evaluate the classication and characterization ability of the different
spectroscopic and spectrometric methods. ANOVA Simultaneous Component Analysis
(ASCA) was applied to achieve a separation of different sources of variance in the complex
sample set. Since the chosen methods and sample preparations probe different parts
and/or molecular constituents of the pollen grains, complementary information about the
chemical composition of the pollen can be obtained. By using consensus principal
component analysis (CPCA), data from the different methods are linked together. This
enables an investigation of the underlying global information, since complementary
chemical data are combined. The molecular information from four spectroscopies was
combined with phenotypical information gathered from the parent plants, thereby helping
to potentially link pollen chemistry to other biotic and abiotic parameters.
Keywords: pollen, consensus principal component analysis, ANOVA simultaneous component analysis, Fourier-
transform infrared spectroscopy, matrix assisted laser desorption/ionization mass spectrometry, surface-
enhanced Raman scattering, Raman spectroscopy, Poa alpina
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17881
Edited by:
Lisbeth Garbrecht Thygesen,
University of Copenhagen,
Denmark
Reviewed by:
Wesley Toby Fraser,
Oxford Brookes University,
United Kingdom
Åsmund Rinnan,
University of Copenhagen,
Denmark
Anna De Juan,
University of Barcelona, Spain
*Correspondence:
Janina Kneipp
janina.kneipp@chemie.hu-berlin.de
Specialty section:
This article was submitted to
Technical Advances in Plant Science,
a section of the journal
Frontiers in Plant Science
Received: 30 August 2019
Accepted: 20 December 2019
Published: 31 January 2020
Citation:
Diehn S, Zimmermann B, Tantseva V,
Seifert S, Bag˘cıog˘lu M, Ohlson M,
Weidner S, Fjellheim S, Kohler A and
Kneipp J (2020) Combining Chemical
Information From Grass Pollen in
Multimodal Characterization.
Front. Plant Sci. 10:1788.
doi: 10.3389/fpls.2019.01788
ORIGINAL RESEARCH
published: 31 January 2020
doi: 10.3389/fpls.2019.01788
INTRODUCTION
The analysis of pollen samples is a crucial task that is necessary in
several elds, including agriculture, plant physiology, ecology,
allergology, and climate studies. Therefore, signicant efforts
have been undertaken to utilize analytical techniques that give
insight into pollen chemical composition, to achieve a
characterization that is more detailed than the morphological
typing by light microscopy.
Vibrational spectroscopic methods, such as FTIR (Pappas
et al., 2003;Gottardini et al., 2007;Dell'Anna et al., 2009;Julier
et al., 2016;Depciuch et al., 2018;Jardine et al., 2019), Raman
scattering (Ivleva et al., 2005;Schulte et al., 2008), and surface
enhanced Raman scattering (SERS) (Sengupta et al., 2005;Seifert
et al., 2016), as well as mass spectrometric methods (Krause et al.,
2012;Lauer et al., 2018) can be applied to classify pollen
according to taxonomic relationships based on molecular
composition. Pollen spectra can also indicate changes in
chemical composition according to genetic background and
environmental inuences (Zimmermann and Kohler, 2014;
Zimmermann et al., 2017;Diehn et al., 2018). A vibrational or
mass spectrum carries ngerprint-like information from all
biomolecular species in the pollen samples that are probed
with the respective spectroscopy, albeit with different selectivity
and sensitivity (Bagcioglu et al., 2015;Diehn et al., 2018). For
example, FTIR spectra of pollen reveal different biochemical
composition for different plant species (Pappas et al., 2003;
Gottardini et al., 2007;Dell'Anna et al., 2009;Zimmermann,
2010;Julier et al., 2016) and within a specic species
(Zimmermann et al., 2017), mainly based on vibrations of
protein and lipid molecules contained in the pollen grains.
Raman microspectroscopy, since based on different selection
rules, can give molecular and structural information
complementary to infrared spectroscopy. Moreover, due to the
different geometry in Raman micospectroscopic experiments and
the penetration depth of the light used to excite the Raman
scattering, different parts of the pollen grains are probed. For
example, Raman spectra show many contributions by stored
starch and lipid bodies and by the sporopollenin polymer that
comprises the pollen exine (Schulte et al., 2008). This biopolymer
consists of coniferyl aldehyde and ferulic acid blocks (Rozema
et al., 2001;Blokker et al., 2006;Li et al., 2019) and provides high
stability and protection to the gametes. Surface-enhanced Raman
scattering (SERS), in turn, gives very strong signals from pollen
constituents that must interact with metal nanoparticles, i.e., the
SERS substrate, and although it enables the investigation of less
abundant molecular species, it has a high selectivity for specic
classes of molecules. This can be the water-soluble pollen
fraction, extracted in a facile way (Seifert et al., 2016) or the
sporopollenin polymer after embedding the SERS nanoparticle
substrate inside the nanoscopic cavities of the pollen shell
(Joseph et al., 2011).
In contrast to vibrational spectra, the molecular basis of
matrix-assisted laser desorption/ionization time-of-ight mass
spectrometry (MALDI-TOF MS) mass spectra from the complex
pollen samples is still much less understood (Krause et al., 2012),
but it was successfully shown to serve as ngerprint-like data for
the classication and identication of pollen species as well
(Krause et al., 2012;Lauer et al., 2018), even at the sub-species
level (Diehn et al., 2018). MALDI-TOF-MS delivers chemical
identiers that are complementary to those of FTIR spectroscopy
when applied to the same set of pollen samples (Zimmermann
et al., 2017;Diehn et al., 2018).
Biospectroscopic data are usually evaluated using multivariate
methods, including principal component analysis (PCA) (Lasch
and Naumann, 1998;Ellis and Goodacre, 2006;Qian et al., 2008).
It factorizes the data matrix that contains all spectra to one score
value for each spectrum and one loading vector for all spectra.
The weighting of the data based on variance in a PCA enables
easier identication of differences in a spectral data set and helps
identication of latent structures (Pearson, 1901;Hotelling, 1933;
Bro and Smilde, 2014). The outcome of a PCA can be explored
easily by scores plots and interpretation of spectral features in
corresponding loadings.
Since each of the four analytical approaches provides unique
information about one particular fraction of the complex pollen
chemistry, a combination of the data in one extensive analysis
would be very promising to improve pollen characterization and
classication. In particular, the combination of different
chemical data is expected to reveal chemical aspects of plant/
pollen phenotype in a more sensitive and more comprehensive
fashion, enabling more insight into, e.g., the adaptation of plants
to environmental conditions. Recent studies show the great
potential of applying consensus principal component analysis
(CPCA) (Wold et al., 1987;Westerhuis et al., 1998) as a multi-
block method to the data from different analytical techniques.
The combination of very different data blocks can be used in the
investigation of biological samples (Perisic et al., 2013), including
pollen (Bagcioglu et al., 2015). CPCA is an extension of the PCA
concept and aims for the maximization of common variation
patterns in the different data blocks. In CPCA, the data blocks are
deated with respect to the variation that is expressed in the so-
called global scores. A difference between PCA on every single
block and a CPCA analysis is that the same variation appears in
the same components in every data block. Thereby, we can
compare results directly between the different types of
spectroscopic information. A potential co-variation in the
blocks is specied in the explained variance of the respective
block scores. Furthermore, a correlation loadings plot can be
generated as result of a CPCA. CPCA not only joins the
information from different data blocks in one analysis, it also
enables the evaluation of interactions between the different
blocks (Hassani et al., 2010;Hassani et al., 2013).
Here, we apply CPCA to the pollen data of the four
complementary methods FTIR spectroscopy, Raman
microspectroscopy, SERS, and MALDI-TOF MS and
compare the results to those of PCA of each of the single data
blocks. The data are measured from pollen samples obtained in
a large-scale greenhouse experiment that was aiming for a
diverse range of investigations connecting to pollen research
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17882
(Zimmermann et al., 2017;Diehn et al., 2018), making also
other phenotypic data on the parent plants available to be
included in the analysis. The sample set discussed comprises
pollen from one grass species, Poa alpina. The parent plants
originate from three different populations, within which four
different growth conditions were applied to individuals of
identical genetic constitution (Figure 1). The design of this
experiment generates two initially separate questions. The rst
is regarding the different chemical composition of pollen from
different populations in the same species. The second question
relates to the differences in pollen composition as a result of
different growth conditions of genetically identical plants within
one population. Therefore, the results of both CPCA and the
PCA are compared for the different spectroscopic methods
and separately for the different design factors, that is,
population and growth condition. One of the aims is to assess
the sensitivity of the multimodal characterization towards
an inuence of population and environmental conditions,
respectively, on pollen chemistry, regardless of the hierarchical
structure of the variation introduced in the specic sample
set. To address this, we have used an ANOVA Simultaneous
Component Analysis (ASCA) in order to investigate the
possibility of separating between different sources of variation
in the complex sample set.
EXPERIMENTAL
Pollen Samples
In a greenhouse experiment, plants of the grass species Poa
alpina were grown under different environmental conditions
using seeds acquired from the Nordic Gene Bank. The seeds
belonged to three different populations of origin, Sweden, Italy,
and Norway, that were chosen to cover geographic and climatic
variation (Figure 1). Details of the growth experiment can be
found in Zimmermann et al. (2017). Briey, for each population,
six individuals were grown from seeds in the spring, and, after
the summer, each individual was divided into four clones. The
plants were subsequently vernalized for 12 weeks at 4°C with a
day length of 8 h. After vernalization, the plants were grown
under long day conditions (20 h), and the respective clones of the
individuals were subjected to four different environmental
conditions: at 14°C and additional nutrients in the irrigation
water (+nu), at 14°C without additional nutrients in the
irrigation water (-nu), at 20°C +nu, and at 20°C nu,
respectively. Pollen samples were collected from the pollinating
plants. Thereby, the sample set contained 24 different pollen
samples for each of the three populations, and the whole sample
set consisted of 72 pollen samples (Figure 1). The pollen grains
were stored at -20°C after collection until further preparation.
FTIR Spectroscopy
Bulk samples of pollen were prepared as homogenous
suspensions and measured by using high-throughput FTIR
accessory. Approximately 1 mg of a pollen sample was
transferred into 1.5 ml microcentrifuge tube containing 500 ml
of distilled water. The sample was sonicated in ice bath, by a 2
mm probe coupled to a Q55 Sonicator ultrasonic processor
(QSonica, LLC, USA) under 100% power. The sonication
period was 2 min in total, with 30 s intermission after the rst
minute of sonication to minimize the increase in temperature.
Following the sonication, the sample suspension was centrifuged
with 13,000 rpm for 10 min, and the suspension was
concentrated by removing 400 ml of supernatant. Of the
remaining suspension, three aliquots (technical replicates),
each containing 8 ml, were transferred onto an IR-transparent
silicon 384-well microtiter plate (Bruker Optik GmbH,
Germany). The microtiter plate was dried at room temperature
for 1 h to create adequate lms for FTIR measurements.
FTIR measurements were obtained using a HTS-XT
extension unit coupled to a TENSOR 27 spectrometer (both
Bruker Optik GmbH, Germany). The system is equipped with a
globar mid-IR source and a DTGS detector. The spectra were
recorded in transmission mode, with a spectral resolution of 4
cm
1
and digital spacing of 0.964 cm
1
. Background (reference)
spectra of an empty well on a microtiter plate were recorded
before each sample well measurement. The spectra were
measured in the 4,000500 cm
1
spectral range, with 32 scans
for both background and sample spectra, and using an aperture
of 5.0 mm. Data acquisition and instrument control were carried
out using the OPUS/LAB software (Bruker Optik GmbH,
Germany). Spectra were pre-processed, rst by taking the
second derivative employing the SavitzkyGolay algorithm
FIGURE 1 | Schematic presentation of the numbers of samples
(corresponding to the amount of analyzed spectra) for populations and
growth conditions. This results in three (Sweden, Italy, Norway) and four (4°C
and additional nutrients, 14°C without additional nutrients, 20°C and
additional nutrients, 20°C and without additional nutrients) group variables for
the two design factors populationand (growth) conditions, respectively.
Abbreviations: +nu, additional nutrients, nu, no additional nutrients.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17883
(Savitzky and Golay, 1964) with a polynomial of degree two and
a window size of 7 points, and second by using extended
multiplicative signal correction (EMSC) with linear and
quadratic components (Martens and Stark, 1991)(Savitzky and
Golay, 1964;Zimmermann and Kohler, 2013). The spectral
range from 8001,800 cm
-1
was used for multivariate analysis.
An average spectrum was calculated from the spectra of the three
technical replicates (aliquots) per sample, resulting in a set of 72
average spectra that were used for further analysis.
Raman Microspectroscopy
Single pollen grains of each pollen sample were measured using a
Raman microspectrometer (Horiba, Bensheim, Germany) with a
50x microscope objective (Olympus, Hamburg, Germany) and a
diode laser operating at a wavelength of 785 nm and an intensity
of 7 · 10
5
W/cm
2
. For each sample, ten spectra from ten different
single pollen grains were collected, using an accumulation time
of 10 s per spectrum. In total, 720 individual spectra were
obtained. Spectral resolution was 1.3-1.6 cm
-1
, considering the
full spectral range. For frequency calibration, six bands in the
spectrum of 4-acetaminophenol (1648.4, 1323.9, 1168.5, 857.9,
651.6, 390.9 cm
-1
) were used. After spike removal, the raw
spectra were interpolated in the range from 400 to 1,750 cm
-1
to achieve an equal distribution of data points across the whole
spectral range. A distance of 1.45 cm
-1
, corresponding to the
average spectral resolution in the experiment was chosen as
distance between variables. Subsequently, a baseline for each
spectrum was estimated by asymmetric least square smoothing
(Eilers, 2003) and subtracted from the respective spectrum,
followed by vector normalization of the baseline corrected
spectrum. An average spectrum was calculated for each sample
from the 10 respective spectra, resulting in a set of 72 average
spectra that were used for further analysis.
Surface-Enhanced Raman Scattering
(SERS)
In the SERS experiments, the water-soluble components of the
pollen grains were extracted and mixed with an aqueous solution
of citrate-stabilized gold nanoparticles as described previously in
reference (Seifert et al., 2016). For this purpose, 100 µl Millipore
water were added to 0.2 mg of the pollen sample. After 5 min, the
samples were centrifuged and the supernatant was pipetted off. 2
µl of this aqueous pollen extract were mixed with 20 µl citrate-
stabilized gold nanoparticles obtained based on the protocol
described in ref. (Lee and Meisel, 1982) and 2 µl of a 0.1 M
sodium chloride solution were added. Subsequently, 20 µl of this
mixture were transferred to a calcium uoride slide for the SERS
measurement. The SERS experiments were performed on a
Raman microscope (Horiba, Bensheim, Germany) in the focal
volume of a 60x water immersion objective (Olympus,
Hamburg) with a laser operating at a wavelength of 785 nm
and an intensity of 2.9 · 10
5
W/cm
2
. Two extracts for each sample
(technical replicates) were prepared and analyzed. For each
extract, 1,000 spectra with an accumulation time of 1 s per
spectrum were collected. This procedure yielded SERS data sets
containing 144,000 individual spectra in total (2,000 spectra per
pollen sample). The spectra were frequency calibrated using a
spectrum of 4-acetamidophenol. Further pre-processing
included spike removal, interpolation, baseline correction, and
vector normalization as described in the previous section. The
2,000 spectra for each sample (obtained from different extracts)
were averaged so that in total 72 average SERS spectra
were analyzed.
MALDI-TOF MS
For the MALDI-TOF MS experiments, each pollen sample was
deposited on a MALDI stainless steel target. 1 µl of formic acid
(90%) was added, and after drying at room temperature, 1 µl of
matrix solution (10 mg of a-cyano-4-hydroxycinnamic acid in 1
ml 1:1 acetonitrile/water and 0.1% triuoroacetic acid) was
applied. (Seifert et al., 2015) MALDI-spectra were obtained in
the mass range from m/z 1,000 to 15,000 using an Autoex III
MALDI-TOF mass spectrometer (Bruker Daltonik, Bremen
Germany) equipped with a 355 nm Smartbeam laser (200 Hz)
and operating in positive linear mode at an acceleration voltage
of 19.13 kV. Two technical replicates for each sample were
prepared, resulting in 144 spectra in total. To obtain equal
distances between the variables, the spectra were interpolated
with a distance of m/z 2 between data points in the mass range
from m/z 5,000 to 9,000, and a 6-degree polynomial baseline
correction was applied before the spectra were vector-
normalized. The two technical replicates were averaged to yield
one spectrum for each sample in order to obtain the same
amount of spectra as in each of the other data blocks.
Morphological and Dry Weight
Measurements of Parent Plants
During the pollination stage, the height of the owering shoots of
the parent plants was determined, using the average value for
three highest owering shoots per individual plant. Furthermore,
the number of owering shoots for each individual was
determined. Plant dry mass was determined at the end of seed
production life stage by cutting the parent plants at ground level
and drying them at 60°C for 24 h.
Chlorophyll Content of Parent Plants
Chlorophyll a and b concentrations were measured by
ultraviolet-visible absorption measurements (Solhaug, 1991)
Leaf samples from each individual were collected during the
pollination life stage. Approximately 48 mg of a sample were
transferred to microcentrifuge tubes containing 1.5 ml N, N-
dimethylformamide, and kept at +4°C for 24 h. The extracts were
measured on a UV-Vis spectrophotometer (Shimadzu 1800,
Japan) using cuvettes with 1 cm path length. Chlorophyll a
and b concentrations were calculated by using absorbance values
at 647 nm and 664 nm according to the equations by Porra et al.
(Porra et al., 1989). The chlorophyll content, morphological and
dry weight data were combined in a separate, fth data block,
termed here additional plant data. The data were normalized
based on data dispersion (autoscaling) before further analysis.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17884
Data Analysis
All pre-processing steps of the spectra and all analyses were
performed using the statistics and machine learning toolbox in
Matlab R2016a (The Mathworks, Inc., Natick, MA, USA). First,
the individual data sets were analyzed by principal component
analysis (PCA) as described in previous work (Seifert et al., 2015;
Seifert et al., 2016;Joester et al., 2017;Zimmermann et al., 2017;
Diehn et al., 2018). Consensus principal component analysis
(CPCA) was used to combine the ve different kinds of data
(FTIR, Raman, SERS, and MALDI-TOF MS spectra, as well as a
block with the additional data obtained from the parent plants),
each one pre-processed as described before. CPCA enables the
analysis of the sample variances within the data blocks and
between different data blocks (Hassani et al., 2010). In order to
apply CPCA, all the data sets need to have the same sample
dimension, and the order of the samples should be identical for
all data sets included in the analysis. Therefore, all technical
replicates were averaged, such that 72 spectra for each data block
were obtained. Outliers were not removed from the averaged
data sets, since each method probes different parts of the
samples, and the data acquisition differs greatly. For the
common representation of the loadings of all the different
kinds of data in one correlation loadings plot as result,
thresholds were dened for the respective data types and those
positive/negative peaks above/below the respective threshold are
represented in the unied plot. For clarity, all other spectral
variables are not shown in these plots. Only the variables
belonging to the additional plant data are displayed as a whole.
To evaluate the ability to discriminate each of the three
groups in the design factor populationas well as the four
individual groups of the design factor growth conditionin the
PCA, Kruskal-Wallis H-test and MANOVA were used. The
populations Sweden, Italy and Norway, as well as the
conditions 14°C and additional nutrients (14+nu), 14°C
without additional nutrients (14-nu), 20°C and additional
nutrients (20+nu) and 20°C and without additional nutrients
(20-nu) are dened as group variables of the design factors
populationand growth condition, respectively (Figure 1).
Kruskal-Wallis H-test and MANOVA were also applied to
the groups of score values after CPCA on the global and block
scores, respectively. As result, the two tests give one p-value and
one d-value for each PCA or CPCA. The Kruskal-Wallis H-test
as a non-parametric statistical test was chosen after assessment of
the data sets regarding their normal distribution. As known from
previous work, the data obtained from SERS experiments do
often not show normal distribution (Seifert et al., 2016). Here,
the test is used to prove the null hypothesis that the distribution
of the data within each respective group, that is, three groups for
the three different populations in the whole data set, and four
groups for the four different growth conditions within each of the
populations, is equal. A p-value below 0.05 indicates a signicant
difference in this distribution for at least one of the groups. The
Kruskal-Wallis H-test was applied to the score values of each
PCA and CPCA that was conducted using the kruskalwallis
function in Matlab. Each of the rst ten components was
investigated, and the p-value was reported using always the
rst PC. However, in case of a high p-value when using the
rst PC, the lowest p-value with any of the other rst ten PCs is
discussed (see Tables 1 and 2). The distribution of the score
values of the rst PC is also visualized using box plots.
MANOVA, comparing the multivariate means for a specic
dimensionality, was executed using the manova1 function in
Matlab. The rst ten PCs (covering at least ~90% of the variance,
with over ~96% of the variance in the FTIR and Raman data sets)
were used for MANOVA, since a balance had to be found
between the requirement to have as much variance as possible
covered, an equal treatment of all data sets, and a reasonable time
for computation. MANOVA was used to estimate the non-
random variation of the group mean of each population and
each growth condition, respectively. In this case, the
dimensionality was either three, corresponding to the three
different populations, or four, due to formation of four groups
corresponding to the four different growth conditions. If the
group means were equal, that is, when no discrimination was
found, the d-value would be 0. A d-value of 1 would indicate that
the group means show a linear dependence on each other, so that
two groups are separated. For the data sets here, the d-value
could reach two in the case of the three different populations and
three in the case of the four different growth conditions.
To calculate the variation in the data induced by the different
design factors, such as population, nutrients, temperature, as well
as their interaction, we used an approach underlying ANOVA-
TABLE 1 | Results of the PCA (p- and d-values) for the discrimination of pollen
samples from all populations and from the individual populations grown under
different environmental conditions.
Method Population p-values for the
separation of the
pollen samples based
on environmental
conditions
d-values for
grouping based
onenvironmental
conditions (max.
3)
FTIR Sweden <0.01 3
Italy 0.035 3
Norway 0.072 (0.011, PC4) 2
All 0.64 (<0.01, PC6) 3
Raman Sweden 0.51 (<0.01, PC4) 1
Italy <0.01 3
Norway 0.36 (<0.01, PC3) 2
All <0.01 2
SERS Sweden 0.73 (<0.01, PC2) 2
Italy 0.37 (<0.01, PC4) 2
Norway 0.64 * 0
All 0.78 (0.014, PC7) 0
MALDI Sweden 0.62 (0.046, PC3) 2
Italy 0.012 1
Norway 0.018 1
All 0.98 (<0.01, PC5) 1
Additional plant data Sweden <0.01 1
Italy <0.01 2
Norway <0.01 2
All <0.01 2
*no p-value below 0.05 can be found for the rst 10 PCs. The p-values are obtained for the
score values of PC1. In case of p-values above 0.05 in PC1, the lowest p-value with any of
the other rst ten PCs and respective PC are shown in parentheses. For the calculation of
d-values, the score values of the rst 10 PCs were used.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17885
PCA and ASCA (Harrington et al., 2005;Smilde et al., 2005),
which are widely used for this purpose (Jansen et al., 2005;de
Haan et al., 2007). In both of these methods an ANOVA model is
established. It represents the original data as a sum of matrices,
each of which corresponds to one design factor. Each of these
matrices consists of the means of the spectra that correspond to
different levels of the factor. As an example, if one design factor
has two levels, its respective matrix will have repeated means of
the two levels in the corresponding rows.
In the ANOVA model used in this study, the design factor
temperaturehas two levels (14°C and 20°C), the design factor
nutrientsconsists of two levels (+nu and -nu), the design factor
interactionis the interaction of temperature' and nutrients' and
has four levels, the design factor populationhas three levels (Italy,
Norway, and Sweden), the factor individual' has 72 levels that
correspond to each individual plant in the set of samples. The
residual variance is summarized in the factor residuals.
ASCA uses the ANOVA model to study the effects of the
design factors on the variation in the data and runs PCA on each
of the matrices to interpret this variation. In ANOVA-PCA the
ANOVA model is analyzed further by PCA to nd if the
differences in the levels for each factor are signicant. In this
study, we use the ANOVA model purely to calculate each design
factor contribution into variation in the data. Since it was of
interest to learn about the variation contributions in each block
of data representing each measurement (FTIR, MALDI, Raman,
SERS and other parental plant data, respectively), the same
analysis was done separately on each data block.
RESULTS AND DISCUSSION
Analysis of the Separate Data Blocks
The well-dened sample set (Figure 1) was measured by the four
different methods FTIR, Raman, and SERS spectroscopy and
MALDI-TOF mass spectrometry, after different pre-processing
of the samples according to the requirement of each spectroscopy
(see Experimental section), leading to the probing of
complementary constituents of the pollen. The four different
types of spectra were obtained from the 72 pollen samples,
constituting four separate data blocks. Furthermore,
phenotypic information from the respective parent plants was
combined in a fth data block.
Figure 2 shows the four types of spectra obtained for the three
different populations, with averaging information in each
population over pollen samples obtained from all parent plants
grown at thefour different environmental conditions.The signals in
the FTIR spectra (Figure 2A) can mainly be assigned to proteins,
represented, e.g., by the amide I and amide II bands at 1669 and
1540 cm
-1
, respectively, to lipids, exemplied by vibrations at 1156,
1467, and 1744 cm
-1
and to sporopollenin, e.g., at 835, 1512, and
1624 cm
-1
, in agreement with spectra reported previously
(Bagcioglu et al., 2015;Zimmermann et al., 2017).
The average Raman spectra in Figure 2B are very similar to
each other as well and display similar signals, albeit at slightly
varying positions, suggesting small differences in the chemical
composition of pollen from different populations. The bands at
1008, 1161, and 1528 cm
-1
can be assigned to carotenoids
(Schulte et al., 2009), while the signals at 526, 549, 725, 855,
1271, 1457, and 1662 cm
-1
are due to vibrations of proteins
(Schulte et al., 2008;Joester et al., 2017). The bands at 483, 1082,
and 1322 cm
-1
are assigned to carbohydrates (Schulte et al., 2008;
Pigorsch, 2009) that can occur at high local concentrations in the
pollen grains as starch deposits. Due to superposition of several
molecular vibrations, some bands in the Raman spectra of pollen
can be assigned to other origins as well. As examples, the bands
at 1161, 1271, 1313, and 1608 cm
-1
could also be assigned to the
ferulic acid and coumaric acid building blocks in sporopollenin
(Blokker et al., 2006;Bagcioglu et al., 2015). Furthermore, the
band at 1608 cm
-1
has also been associated with mitochondrial
activity (Huang et al., 2004;Pully and Otto, 2009).
Due to the sample preparation as aqueous extract and the use
of aqueous nanoparticle solutions, the SERS experiments probe
the water-soluble fraction of the pollen grains. Because of the high
variation in the SERS spectra caused by the specics of the SERS
experiment, high numbers of spectra are needed for a reliable
statistical analysis. (Seifert et al., 2016) Therefore, 2,000 spectra
were measured from each sample, resulting in reproducible
average spectra that are based on 24,000 individual spectra per
population. They are shown in Figure 2C. The average spectra
TABLE 2 | Results of the CPCA (p- and d-values) for the discrimination of pollen
samples from all populations and from the individual populations grown under
different environmental conditions.
Method Population p-values for the
separation of the
pollen samples based
on environmental
conditions
d-values for
grouping based
onenvironmental
conditions (max.
3)
Global Sweden 0.036 2
Italy <0.01 3
Norway <0.01 2
All 0.95 (<0.01, CPC3) 3
FTIR Sweden 0.032 2
Italy 0.013 3
Norway 0.084 (0.021, CPC4) 3
All 0.70 (0.011, CPC3) 3
Raman Sweden 0.47 (0.013, CPC6) 2
Italy <0.01 3
Norway 0.19 (<0.01, CPC5) 3
All 0.64 (<0.01, CPC3) 3
SERS Sweden 0.093 (0.032, CPC5) 2
Italy 0.089 (<0.01, CPC3) 3
Norway 0.40 * 2
All 0.98 (<0.01, CPC3) 3
MALDI Sweden 0.60 (0.027, CPC5) 2
Italy <0.01 3
Norway <0.01 3
All 0.98 (<0.01, CPC3) 2
Additional plant data Sweden <0.01 2
Italy <0.01 3
Norway <0.01 2
All <0.01 3
*no p-value below 0.05 can be found for the rst 10 PCs. The p-values are obtained for the
score values of PC 1. In case of p-values above 0.05 in PC 1, the lowest p-value with any
of the other rst ten PCs and respective PC are shown in parentheses. For the calculation
of d-values, the score values of the rst ten PCs were used.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17886
are very similar, and their characteristic bands can mainly
be assigned to vibrational modes of nucleobases, e.g., at 494,
649, 735, 802, 921 cm
-1
(Seifert et al., 2016) and amino acids, at
995, 1021, 1221 cm
-1
(Kyu et al., 1987;Stewart and Fredericks,
1999), in agreement with the probing of water-soluble
biomolecules extracted from the pollen.
MALDI TOF mass spectrometry was utilized to detect large
molecules with a mass over 5 k Dalton. Figure 2D displays
average MALDI-TOF mass spectra showing peaks at m/z 5282,
5322, 5404, 5580, 5718, 5853, 5980, 6128, 6296, 6776, and 8136.
The differences in the population averages are obvious and
indicate that the pollen samples differ in their composition in
each population. From earlier attempts to interpret the bands we
infer that they include oligosaccharides (Krause et al., 2012;
Seifert et al., 2015;Diehn et al., 2018) and larger peptides.
By PCA of the respective type of spectra/data, the pollen
samples of the three different populations can be discriminated
using each of the individual data blocks. To visualize the
distribution of the score values of PC1 for each population,
Figure 3 shows the corresponding box plots with minimum and
maximum score values. Outliers are mainly observed for the
SERS data (Figure 3C) due to high variation in this type of data
owing to the specic measurement approach (Seifert et al., 2016).
P-values below 0.05 for all ve data sets indicate a separation of
at least one group for all these data sets.
The box plotsof Figure 3 show that an unequivocal separation of
all three populations based on PC1 is only possible when the
MALDI-TOF MS scores (Figure 3D) are used. The scores of
FTIR (Figure 3A) and Raman data (Figure 3B) for example
FIGURE 2 | Continued
FIGURE 2 | FTIR (A), Raman (B), SERS (C), and MALDI-TOF MS (D)
spectra of pollen from the populations Sweden (black), Italy (blue), and
Norway (red). All spectra were pre-processed according the requirements of
the respective spectroscopic method and are averages from the respective
population, including samples obtained for all growth conditions. The spectra
are stacked for clarity. FTIR, Fourier-transform infrared spectroscopy; Raman,
Raman spectroscopy; SERS, surface enhanced Raman scattering; MALDI-
TOF MS, matrix assisted laser desorption/ionization mass spectrometry.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17887
show very similar distributions for the two populations Sweden and
Italy. In order to includemore than one principal component when
evaluating separation of the three populations by PCA, d-values
were determined by MANOVA of the scores of the rst ten
principal components of each PCA/data block. For all data blocks
a d-value of 2 is obtained. This indicates the separation between
three groups, corresponding here to the three populations.
Therefore, we conclude that a separation of the three different
populations is possible with any of the data sets.
The parent plants in each population were grown under four
different conditions. Discrimination regarding potential effects of
additional nutrients and temperature as design factors on pollen
chemistry was studied using each of the ve data sets separately
as well. This was done for each population individually, as well as
for all populations together. Table 1 summarizes for each data
block the PCA results. The p-values were determined using one
PC (Table 1, left column). In case of a high p-value when using
the rst PC, the lowest p-value with any of the other rst ten PCs
is shown in the table. The d-values were determined using the
rst ten principal components (Table 1, right column).
The rst section of Table 1 displays the outcome of the PCAs
obtained from the FTIR data sets. The separation based on FTIR
spectra receives a p-value below 0.05 and a d-value of 3 for the
populations Sweden and Italy, indicating that FTIR data alone
enable differentiation of the applied growth conditions for these
two populations (compare also the box plots in Figures S2 and
S3). The FTIR data set of the population Norway with p-value
larger than 0.05 and a d-value of 2 comprises less variance
between growth conditions. When all populations are analyzed
together, a high p-value for the rst PC is obtained, which means
that none of the four different growth conditions is separated
FIGURE 3 | Continued
FIGURE 3 | Results of the principal component analysis of FTIR (A), Raman
(B), SERS (C), MALDI-TOF MS (D), and additional plant data (E),
respectively. Box plots display the variation of the score values of the rst PC
regarding each population obtained by Kruskal-Wallis H-Test. Red lines
indicate the median of the respective distribution, blue boxes represent the
interquartile range, the black lines demarcate minimum and maximum values,
and extreme values are shown as red markers. FTIR, Fourier-transform
infrared spectroscopy; Raman, Raman spectroscopy; SERS, surface
enhanced Raman scattering; MALDI-TOF MS, matrix assisted laser
desorption/ionization mass spectrometry.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17888
using the variance explained by the rst PC. Nevertheless, using
the 1st to 10th PC, the d-value of 3 indicates a possible separation
of all four growth conditions by FTIR alone.
Using the Raman data sets, the p and d-values of the PCAs
from the data of the populations Sweden and Norway indicate a
less sufcient discrimination ability (Table 1, second section).
Only for the population Italy, a low p-value and a d-value of 3
can be interpreted as a separation of the four groups of the
different growth conditions. In addition, the analysis of all
populations together leads to a small p-value, showing the
separation of at least one group based on Raman spectral
information. We attribute the smaller discrimination ability
compared to FTIR to this different selectivity of Raman
spectroscopy. The high variances according to the growth
conditions in the population Italy explained by the rst PC are
remarkable, and in good agreement with previous studies on
phenotypic plasticity in pollen (Zimmermann et al., 2017). The
higherthephenotypicplasticity,themorethechemical
composition in pollen varies when environmental conditions
change. The high phenotypic plasticity of the population Italy
has been inferred from MALDI TOF MS and FTIR spectra of the
same Poa alpina population previously (Zimmermann et al.,
2017;Diehn et al., 2018), where a lower inner-group variance
regarding different genotypes of the plants was found.
Investigation of the SERS spectra from aqueous pollen extract
by PCA results in p-values above 0.05 for each individual
population as well as the whole data set (Table 1, third section),
clearly showing that an analysis of the samples by SERS alone will
not be sufcient for the discrimination of pollen from parent
plants that were grown under different environmental conditions.
Nevertheless, according to the p-values found in PC2 in
population Sweden and PC4 in population Italy (p-values in
parentheses in Table 1), the variances from the effect of the
growth conditions can also be detected in the aqueous extract for
these two data sets and therefore add complementary
information in the multi-block analysis discussed below.
The MALDI TOF MS data from population Sweden lead to a
p-value above 0.05, whereas the p-values for the other two
populations stay slightly below 0.05 (Table 1, fourth section).
In contrast, based on the outcome for the whole population, we
cannot conclude chemical variance as result of different growth
conditions using only one PC. The d-value of 1 (obtained using
the rst ten PCs), found for the whole data set as well as for
population Italy and population Norway, can be interpreted as
the formation of two distinct groups of MALDI spectra. This is in
agreement with our previous results (Diehn et al., 2018), where
we found a high ability to discriminate between pollen from
plants growing with additional nutrients and pollen from plants
without additional nutrients using MALDI data from the same
samples of Poa. Since the discrimination takes place in the range
m/z 5000-9000, we infer that the detected signals belong to
proteins and their derivatives from pollen nutrient storage.
The last section in Table 1 contains the p- and d-values for
the analysis of additional plant phenotype data, namely height
and number of owering shoots, plant dry mass, and chlorophyll
content. The p-values for each population and for the data set
with all three populations combined are below 0.05, and we
conclude that the variances regarding the separation of at least
one specic group of scores from the other growth conditions are
high. The d-values for the analyses of the data sets, however, are 2
or smaller, indicating that discrimination regarding all four
growth conditions is not obtained. This is also be illustrated in
the box plots in Figures S1S4 (last rows).
The variation contributions of the different design factors,
such as population, nutrients, temperature and their interaction
as well as the contributions from individual variation, were
calculated with an approach underlying ANOVA-PCA and
ASCA (Harrington et al., 2005;Jansen et al., 2005;Smilde
et al., 2005;de Haan et al., 2007). Figure 4A shows the
contribution of all possible design factors, that is, each type of
variation for the whole data set of 72 spectra for each method.
The variation contribution of the populations (Figure 4A, cyan
bars) is very large in the four spectroscopic/spectrometric data
sets, larger than the variation contribution due to the different
growth conditions (Figure 4A, blue, orange and yellow bars).
Interestingly, and in agreement with previous work
(Zimmermann et al., 2017), the contributions of variation of
the individual samples (Figure 4A rst column, purple bars) is of
similar magnitude as that introduced by changes in growth
condition of the parent plant, and in the data sets from SERS
and MALDI (Figure 4A second and third column, purple bars),
this contribution by individual variation is even larger.
Considering the data gathered from the parent plants, the
largest variation contribution is the effect of the nutrient addition
(Figure 4A, rightmost bar, orange coloring), obviously having
more consequences for the constitution of the plant itself than
for the chemical make-up of the pollen. In addition, differences
between phenotypic features of the plants in the different
populations are of a similar magnitude as variation due to
individual differences. (Figure 4A, rightmost bar, cyan
coloring). The contribution of the residual variation (Figure
4A, green bars) is relatively high for all data sets. In some, such as
Raman and SERS (Figure 4A, second and third column,
respectively), the residual variation contributes the most. We
think that this must be due to the type of experiment, which are
in these cases much more prone to spectrum-to-spectrum
uctuation. Moreover, the Raman and SERS data sets were
collected over a course of several weeks, whereas MALDI and
FTIR were high-throughput measurements obtained in one-
preparation procedures. So, the big residual variation in SERS
and Raman can be explained by the experimental variations.
In Figure 4B, relative variation contributions of the growth
conditions, namely temperature, nutrients and the interaction of
both factors are presented. Variations by these factors are
emphasized by omitting population variation, individual
variation, and residual variation. To calculate these, a variation
of each factor was normalized by the sum of the variations for the
three factors of interest. While the variation contribution of both,
the temperature and nutrients is high for the three spectroscopic
methods FTIR, Raman and SERS, for MALDI the variation of the
nutrient factor is higher than the variation contribution of
the temperature.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 17889
The contribution of the different design factors to the total
variation were also analyzed for each population separately
(Figures 4C, D,Figure S5). As an example, Figures 4C, D
show the outcome of the analysis for the population Italy. For the
population Sweden (Figures S5A, B), the overall variation
contribution of the individuals (Figure S5A, purple), is higher
compared to the other populations, and contribution of variation
due to the growth conditions is rather small.
This type of analysis helps understanding the underlying
variation in the data introduced by different design factors and
by other unwanted factors. PCA analysis and other multivariate
data analysis techniques, if successfully working on the data,
ensure that the amount of relevant variation available in the data
is sufcient to discriminate between groups. As an example,
although the different growth conditions contribute to only 10%
of the variation in the FTIR data from all populations (Figure
4A,rst column), we observed a good discrimination of growth
conditions using the rst ten PCs, yielding a d-value of 3 (Table
1,rst section). This shows that the methods are powerful
enough to focus on the relevant information in the data, and
that the residual variation is not systematic. Regarding the
hierarchical nature of the variance, the results of the ASCA
approach are in good agreement with the results obtained by
PCA. In data sets that show large contributions by different
sources of variation, separation in a PCA is not unequivocal (see
Table 1).
In conclusion, the different analytical methods vary greatly in
their potential todiscriminate the pollen from the sample set based
on population and environmental inuences. For FTIR
spectroscopy (Zimmermann et al., 2017)andMALDI-MS
(Diehn et al., 2018) this has been discussed previously. Due to
the different selectivity in MALDI compared to FTIR, there is a
superposition by the variation between the different genotypes
(that is, individual variation) that impairs the discrimination
ability for different growth conditions within one population
(Diehn et al., 2018). While both Raman micro-spectroscopy of
single pollen grains and SERS enable classication of the pollen
samples with respect to the corresponding population, no strong
FIGURE 4 | (A) Variation contribution of the design factors temperature (blue), nutrients (orange), the interaction of temperature and nutrients (yellow), different
individuals (purple), populations (cyan), and residual variance (green) for the 72 spectra from the whole data set. (B) Relative contribution of temperature (blue),
nutrients (orange), and the interaction of both (yellow) for the 72 spectra from the whole data set (C) and variation contribution of the design factors temperature
(blue), nutrients (orange), the interaction of temperature and nutrients (yellow), individuals (purple) and the residuals (green) for the 24 spectra from the population
Italy. (D) Relative contribution of temperature (blue), nutrients (orange), and the interaction of both (yellow) for the 24 spectra from the population Italy. In (B, D), the
contribution to the variance by specic population and the residual variance were left out, and the variation of each factor was normalized by the sum of the
variations for the three other factors of interest.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178810
variation is found when these data sets are used to assess separation
according to the varied environmental conditions of the parental
plants. Nevertheless, the variation due to varied growth conditions
is highly dependent on the considered population.
CPCA for the Classication of Pollen
Samples According to Plant Populations
With consensus principal component analysis (CPCA), the ve
individual data blocks can be combined, and the impact of each
method on a global analysis can be evaluated. Figure 5 shows the
resultsoftheCPCAfortheclassication of the different
populations of Poa alpina, consisting of ve block score plots
(Figures 5BF) that correspond to the different analyses, and of
a global scores plot (Figure 5A). The global score values of the
rst and second PC (Figure 5A) show a clear discrimination of
the three different populations. In particular, based on the
variance represented by CPC1, data from the population
Norway and data from the population Italy are separated. As
revealed by the block scores plots, the rst component is mostly
inuenced by the FTIR block, comprising 41.7% explained
variance (Figure 5B) and the MALDI block, explaining 39.62%
of the variance (Figure 5E). The second PC is inuenced in
particular by the SERS data, explaining 37.55% of the variance
(Figure 5D) and the block with the data on the parent plants,
explaining 21.84% of the variance (Figure 5F). In all of the scores
plots, the data sets of the population Sweden have positive score
values, while the data sets of the populations Italy and Norway
have mostly negative values regarding CPC2 (Figure 5),
particularly for the Raman (Figure 5C), SERS (Figure 5D),
and MALDI (Figure 5E) block. A CPCA containing FTIR,
Raman, SERS and MALDI without the additional plant
information leads to very similar scores plots, where also all
three populations would be discriminated (Figure S6).
In Figure 6, the results of the separation of the respective rst
CPC are summarized in box plots for each block as well as for the
global scores (Figure 6). Furthermore, we calculated a d-value of
2 based on the CPCA scores of CPC1 to CPC10 for the global
scores as well as for all block scores. The data indicate that
separation of the three populations is readily achieved based on
the global scores (Figure 6A), and that the FTIR (Figure 6B) and
the MALDI data sets (Figure 6E) have the greatest inuence on
the separation in the global scores.
In order to analyze which variables of the respective methods
cause the separation in the global analysis and to investigate the
correlations between them, a correlation loadings plot was
generated (Figure 7). It shows the correlation between the
global scores of the populations Sweden (red cross), Italy (red
circle) and Norway (red triangle) and the relevant variables of the
different blocks. For the clarity only the extrema of the loadings
of the rst and second component from the spectroscopic and
MALDI blocks are shown, as well as all ve variables from the
additional plant data. Therefore, there are no variables visible
close to the origin of the plot. The different populations are
characterized by variables that are located close to the global
scores of the populations. The separation of the data from the
population Sweden is caused by a high amount of spikes in the
FIGURE 5 | Score values of the CPCA analysis for the classication of samples from the populations Sweden (black crosses), Italy (red circles), and Norway (blue
triangles). (A) Scores for the global scores (BF) individual data blocks. CPCA, consensus principal component analysis.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178811
respective progenitor plants and their high dry mass. In addition,
this population is characterized by Raman bands at 1007, 1161,
and 1529 cm
-1
that can be assigned to carotenoids (Schulte et al.,
2009) as well as by bands at 555 cm
-1
that can be assigned to
proteins (Schulte et al., 2008), and a MALDI peak at m/z 6038.
The great inuence that the SERS data block has on CPC2,
separating population Sweden (see Figure 5C), reects in a
correlation with SERS signals at 416, 733, 994, 1154 and 1545
cm
-1
that are particularly important to discriminate the pollen
data from the population Sweden (Figure 7, magenta markers).
In the two other populations, SERS signals at 581, 774, 1051,
1379, and 1424 cm
-1
are observed. They might be attributed to
the water-soluble part of proteins or carbohydrates.
The differentiation between the populations from Norway
and Italy is achieved utilizing CPC1. The population from Italy is
mainly separated by chemical information contained in the FTIR
bands (Figure 7, blue markers) at 1026, 1079, 1151, 1472, 1525,
1649, and 1688 cm
-1
, Raman bands (Figure 7, green markers) at
484, 649, 948, and 1609 cm
-1
, and MALDI TOF MS peaks
(Figure 7, yellow markers) at m/z 5282, 5968, 5980, and 6264.
The FTIR and Raman bands can be assigned to starch, protein
and sporopollenin vibrations (Schulte et al., 2008;Zimmermann,
2010;Bagcioglu et al., 2015). Although an assignment of the
MALDI peaks is more challenging, their positive correlation with
FIGURE 6 | Continued
FIGURE 6 | Results of the consensus principal component analysis of the
ve data sets visualized by box plots of the global scores (A) and the blocks
scores for FTIR (B), Raman (C), SERS (D), MALDI-TOF MS (E), and
additional plant data (F). The box plots display the variation of the score
values of the rst CPC regarding each population obtained by Kruskal-Wallis
H-Test. Red lines indicate the median of the respective distribution, blue
boxes represent the interquartile range, the black lines demarcate minimum
and maximum values, and extreme values are shown as red markers. FTIR,
Fourier-transform infrared spectroscopy; Raman, Raman spectroscopy;
SERS, surface enhanced Raman scattering; MALDI-TOF MS, matrix assisted
laser desorption/ionization mass spectrometry.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178812
these bands suggests that some of them are connected to
nutrients, in agreement with previous discussions suggesting
their assignment to oligosaccharides (Krause et al., 2012;
Seifert et al., 2015;Diehn et al., 2018).
The data sets of the population Norway show a positive
correlation to the FTIR vibrational bands at 1089, 1166, 1503,
1666, and 1746 cm
-1
as well as to the MALDI peaks at m/z 5880
and 6296 (Figure 7, bottom left section). Variances in Raman
bands at 829 and 1043 cm
-1
are positively correlated to the
population Norway. Most of the Raman bands can be assigned as
protein vibrations (De Gelder et al., 2007), whereas the FTIR
bands could be mainly assigned to carbohydrates (Bagcioglu
et al., 2015;Zimmermann et al., 2017).
As illustrated by the band assignments, in addition to a
redundancy in information (e.g., in some bands in FTIR and
Raman spectra) each data block contains some exclusive
molecular information, leading to their complementarity. The
different contribution of the ve data blocks in the discrimination
of the three populations shown in the correlation plot (Figure 7)
indicates that particular parts of the pollen chemistry are
responsible for the differences between populations, and that
very different molecular/compositional parameters are
responsible in the biochemical variation between two
populations. The MALDI-TOF MS data have great inuence on
the analysis and can be exploited for a precise discrimination of all
three populations. This is in accordance with the results of the
PCA of the isolated data block above (Figure 3D) in this paper and
supports previous results that indicate that MALDI-TOF MS and
the biochemical ngerprint of glycoproteins and other
macromolecules are specic for the pollen of a particular grass
population (Diehn et al., 2018).
CPCA for the Classication of Pollen
Samples According to Different
Environmental Inuences
CPCA was applied as well to discriminate between pollen
samples within each population that were collected from
progenitor plants grown under four different environmental
FIGURE 7 | CPCA correlation loadings plot for the 1st and 2nd CPCA component. Displayed are the global scores of the three populations Sweden, (red cross),
Italy (red circle), and Norway (red triangle), as well as the loadings of the blocks of FTIR, Raman SERS, MALDI-TOF and additional plant data. For clarity only extrema
of the loadings were shown for the spectroscopic/spectrometric data. CPCA, consensus principal component analysis; FTIR, Fourier-transform infrared
spectroscopy; Raman, Raman spectroscopy; SERS, surface enhanced Raman scattering; MALDI-TOF MS, matrix assisted laser desorption/ionization mass
spectrometry.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178813
conditions: 14°C and additional nutrients, 14°C without
additional nutrients, 20°C and additional nutrients, 20°C
without additional nutrients. Table 2 shows the resulting p
and d-values analyzing the whole data set from all populations
and the data from each of the three different populations
individually for the global scores (Table 2,rst section) and all
the block scores (second to sixth section, respectively). The p-
values for the global scores are below 0.05 for each population,
indicating the separation of the different groups in the rst
CPCA component (Table 2,rst section). However,
considering all three populations together, separation is based
on the third CPCA component. MANOVA utilizing the rst ten
CPCA components shows the highest possible d-value of 3,
proving successful classication of all four groups of samples
for population Italy, as well as the for the whole sample set of all
three populations. The lower d-value for the global scores in the
population Sweden and Norway may be explained by a lower
phenotypic plasticity of these populations compared to the
population Italy. This is in good agreement with previous
analyses of other data of some of the samples discussed here
(Zimmermann et al., 2017;Diehn et al., 2018).
Comparison of the results for the block scores (Table 2,
second to sixth section) will help to identify those data blocks
that are responsible for a separation based on the global scores.
Based on the d-values, a separation of the samples into four
groupscorresponding to four environmental conditionsis
observed when all populations are analyzed together (last line
in each of the sections of Table 2). The separation into four
groups is possible for each of the ve block scores except those of
the MALDI block (last line in section 5 of Table 2). By
interpreting the corresponding dendrogram shown in Figure
S7, these three groups correspond to the condition 14°C without
additional nutrients, 20°C without additional nutrients, and
plants that obtained additional nutrients (regardless of growth
temperature). The Raman block scores indicate separation of the
four groups in the two populations Norway and Italy (Table 2,
third section). For the other block scores (FTIR, SERS, and
MALDI), the separate analysis of each of the populations gives
very different results, with the samples from the population Italy
showing separation according to the four growth conditions
(Figure 1) in most of them, but less than four distinct groups in
the populations Sweden and Norway. The block scores for the
data gathered from the parent plants show very similar behavior
and result in clear classication of all four conditions only in
population Italy (Table 2, sixth section).
The weighting of each block in the CPCA can be interpreted
and allows more insight into the inuence the data blocks on each
other. As an example, Figure 8 shows the results for the CPCA
applied to the data of the pollen samples from the population
Italy. The rst component of the global scores plot (Figure 8A)
separates between positive scores values of the data of samples
from progenitor plants that were grown with the addition of
nutrients (black crosses and blue triangles) and negative score
values for samples from plants that were grown without the
addition of nutrients (red circles and green diamonds). In Figures
8C, F, the great inuence of the Raman and the additional plant
data block, respectively, are revealed. Both blocks display similar
group formation in the scores plots, with high variances explained
by the rst CPC of 35.52% and 54.50%, respectively. The
corresponding p-values in Table 2 are very low.
The scores of the second CPCA component separate pollen
samples grown at 14 °C, as well as at 20 °C without additional
nutrients (black crosses, red circles and green diamonds) with
positive values from negative values of those pollen samples
grown at 20°C without additional nutrients (blue triangles)
(Figure 8A). The CPC2 is mainly inuenced by the SERS data
(Figure 8D), explaining 33.77% of the variance. In the plot of the
block score values (Figure 8D), no separation of the groups that
could correspond to growth conditions of the plants can be
found. This suggests that other sources of variance, in this
experiment resulting from individual genotypes, superimpose
the inuence of the growth conditions as discussed for other data
previously (Diehn et al., 2018). It is also in agreement with the
calculated p- and d-values for the SERS block (Table 2, section
4). Furthermore, the Raman block scores plot, as well as the
scores from the additional plant data, show great potential
regarding the discrimination of different growth conditions in
the population Italy. Since the additional plant data block
explains most of the variance in the rst CPC, CPCA was also
performed without it, by using only the spectroscopic/
spectrometric data blocks, in order to conrm that the
obtained global pattern is also driven only by the pollen chemical
composition (compare with Figure S8), not by phenotypic features
of the parent plant. Nevertheless, the additional plantdata lead to a
more complete view in this study and show correlation to the
spectroscopic data blocks (compare Figure 9).
The molecular differences that cause the separation of the
data reveal themselves in the correlation loadings plot for the
data from population Italy (Figure 9). Again, only those loadings
with the highest impact are shown for clarity and only the
variables of the additional plant data were presented in full. As
expected after the discussion of the block scores (Figure 8), the
rst CPCA component that separates samples from plants grown
with additional nutrients (crosses and triangles) from samples
without additional nutrients (circles and diamonds, also
compare Figure 8) is mainly inuenced by the Raman block
and the additional plant data.
Raman bands that represent pollen samples with nutrient
addition are 474, 830, 1003, 1435, and 1602 cm
-1
. The bands at
1435 cm
-1
and 1602 cm
-1
can be assigned to lipids (Ivleva et al.,
2005;Schulte et al., 2008) and a high mitochondrial activity,
respectively (Huang et al., 2004;Pully and Otto, 2009). The other
bands are associated with proteins (Schulte et al., 2008;Bagcioglu
et al., 2017). The negative scores of the rst CPC and the data of
the pollen samples without additional nutrients (Figure 9,
diamonds and circles) are mainly inuenced by Raman bands
at 485, 949, 1010, 1138, and 1471 cm
-1
.Thesebandsare
associated with carbohydrates, such as starch (Schulte et al.,
2008;Pigorsch, 2009;Bagcioglu et al., 2017). Pollen are storing
their nutrients in lipid bodies as well as in starch bodies, which
are occupying most of the space in pollen grains (Wang et al.,
2015). Our results conrm that plants growing under different
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178814
nutrient conditions vary in their quality and/ or amount of such
storage bodies inside the pollen.
The second component can be used to separate between rather
positive scores values corresponding to samples that were grown
at low temperatures (crosses and circles) and negative scores
values corresponding to samples that were grown at higher
temperatures (diamonds and triangles). As discussed before
(compare Figure 8), this separation is mainly inuenced by
SERS and FTIR bands. In particular, samples from plants grown
at lower temperatures are characterized by a set of SERS bands
that include 445 cm
-1
and the FTIR bands at 1721 and 1475 cm
-1
.
The FTIR signals can be assigned to lipids (Bagcioglu et al., 2015).
Samples grown under higher temperatures are characterized by
SERS bands at 419, 929, 957, and 1,564 cm
-1
, and FTIR bands at
1,666 and 1,503 cm
-1
. The bands could be assigned to nucleobases
and proteins (Seifert et al., 2016). Based on the inuence the
combination of SERS and FTIR data, we can assume that the
discrimination regarding the different growth conditions is
probably mostly inuenced by the chemical composition of the
pollen interior, although in the preparations for SERS
experiments- also water soluble compounds from the pollen
outer shell may be found in the aqueous extract.
To summarize the results from the correlations loadings plot,
discrimination of different nutrient conditions is mainly
inuenced by Raman bands that can be assigned to pollen
outer shell and nutrient storage, as well as by plant parameters
that are present in the additional plant data block. From our
results, we infer on differences in amount and quality in lipid and
starch bodies inside the pollen grains to be responsible for a
distinction of samples from plants grown at different nutrient
conditions. This is in good agreement with previous studies on
Poa alpina using only FTIR spectroscopy (Zimmermann et al.,
2017). The temperature conditions at which parent plants are
grown mainly affects the SERS and FTIR data blocks and,
probably, mainly the chemical composition of the interior of
the pollen grains. It has to be pointed out that this conclusion is
only made based on the data of the pollen from population Italy,
were the samples are showing the highest phenotypic plasticity of
the three investigated populations. Within the other populations,
the correlation of the signals can differ greatly, indicating higher
phenotypic rigidity, as discussed above.
CONCLUSION
A well described sample set of pollen from Poa alpina was
analyzed by FTIR spectroscopy, Raman microspectroscopy,
surface enhanced Raman scattering (SERS) and MALDI TOF
mass spectrometry, as well as by collected additional data from
the parental plants. The chosen methods are complementary
regarding sample preparations, selectivity, and sensitivity of the
analytical technique. Our results show the ability to detect and
describe variances within the pollen composition related either to
the place of origin of parent plants (i.e., populations) or the
growth conditions. However, suitable data analysis is needed in
order to discuss the relatively small chemical differences in these
complex biological samples.
FIGURE 8 | Scores of the CPCA analysis for the classication of samples from pollen of the population Italy regarding the four different growth conditions 14°C and
additional nutrients (black crosses), 14°C without additional nutrients (red circles), 20°C and additional nutrients (blue triangles), and 20°C without additional nutrients
(green diamonds) for (A) the global scores and (BF) the ve block scores.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178815
The sample set is designed using plants from three
populations that were grown under different nutrient
conditions and temperatures. Therefore, different levels of
classication and inuence on pollen composition could be
analyzed. As expected, the separation of groups in the sample
set according to populations and growth conditions, respectively,
is not achieved equally well by each of the methods. As shown by
an analysis of different sources of variance using ASCA, different
analytical techniques are emphasizing different parameters of
pollen chemistry related either to the genetic background or the
environmental inuence. This has been suggested in previous
work where data from FTIR (Zimmermann et al., 2017) and
MALDI-TOF-MS (Diehn et al., 2018) on a similar sample set
were analyzed but has been shown here using three more, very
different types of data. By combination of the different data
blocks in a CPCA, a complete set of many differences, observed
with the complementary methods can be used to describe the
variation with respect to the different groups. We have also
compared the individual classication ability of the different
methods and the different levels using PCA in combination with
simple statistical tests. The different populations can be easily
distinguished using MALDI-TOF MS, whereas the three
spectroscopic methods are more suitable to separate between
different growth conditions. Moreover, as discussed, the same
data blocks can have a different inuence on the distinction
between different growth conditions in the three populations.
This implies that, due to the different fraction of the pollen
chemistry that is represented by each data block (or analyzed by
each of the methods), the biochemical effect of the growth
conditions on pollen chemistry can vary for different
populations. This would be in agreement with variation in
phenotypic plasticity between the populations, in particular
FIGURE 9 | CPCA Correlation loadings plot for the 1st and 2nd CPCA component. Displayed are the global scores of population Italy regarding the four growth
conditions 14°C and additional nutrients, (black crosses); 14°C without additional nutrients, (red circles); 20°C and additional nutrients (blue triangles); 20°C without
additional nutrients (green diamonds), as well as the loadings of the blocks of FTIR, Raman SERS, MALDI-TOF and additional plant data. For clarity only extrema of
the loadings are shown for the spectroscopic/spectrometric data. CPCA, consensus principal component analysis; FTIR, Fourier-transform infrared spectroscopy;
Raman, Raman spectroscopy; SERS, surface enhanced Raman scattering; MALDI-TOF MS, matrix assisted laser desorption/ionization mass spectrometry.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178816
regarding different metabolic and molecular pathways used in
environmental adaptation.
DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to
the corresponding author.
AUTHOR CONTRIBUTIONS
AK, BZ, JK, MO, and SF conceived the research idea. SF designed
the growth experiments. AK, BZ, and MB designed the
FTIR experiments. JK, SD, and SS designed the Raman and SERS
experiments. JK, SD, SS, and SW designed the mass spectrometry
experiments. BZ and MB performed sampling, FTIR experiments
and the measurement of the additional plant data. SD performed
the mass spectrometry,Raman and SERS experiments. BZ, SD, and
VT analyzed the data. JK and SD wrote the article. AK, BZ, MB,
MO, SF, SS, SW, and VT discussed and revised the article.
FUNDING
The research was supported by the European Commission
through the Seventh Framework Programme (FP7-PEOPLE-
2012-IEF project No. 328289) and ERC Grant No.259432 to JK.
ACKNOWLEDGMENTS
We thank Øyvind Jørgensen (Norwegian University of Life
Sciences) for taking care of the plants. We acknowledge support
by the German Research Foundation (DFG) and the Open Access
Publication Fund of Humboldt-Universität zu Berlin.
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online
at: https://www.frontiersin.org/articles/10.3389/fpls.2019.
01788/full#supplementary-material
REFERENCES
Bagcioglu, M., Zimmermann, B., and Kohler, A. (2015). A multiscale vibrational
spectroscopic approach for identication and biochemical characterization of
pollen. PloS One 10, e0137899. doi: 10.1371/journal.pone.0137899
Bagcioglu, M., Kohler, A., Seifert, S., Kneipp, J., and Zimmermann, B. (2017).
Monitoring of plant-environment interactions by high-throughput FTIR
spectroscopy of pollen. Methods Ecol. Evol. 8, 870880. doi: 10.1111/2041-
210X.12697
Blokker, P., Boelen, P., Broekman, R., and Rozema, J. (2006). The occurrence of p-
coumaric acid and ferulic acid in fossil plant materials and their use as UV-
proxy. Plant Ecol. 182, 197. doi: 10.1007/978-1-4020-4443-4_13
Bro, R., and Smilde, A. K. (2014). Principal component analysis. Anal. Methods 6,
28122831. doi: 10.1039/C3AY41907J
De Gelder, J., De Gussem, K., Vandenabeele, P., and Moens, L. (2007). Reference
database of Raman spectra of biological molecules. J. Raman Spectrosc. 38,
11331147. doi: 10.1002/jrs.1734
De Haan, J. R., Wehrens, R., Bauerschmidt, S., Piek, E., Van Schaik, R. C., and
Buydens, L. M. C. (2007). Interpretation of ANOVA models for microarray
data using PCA. Bioinformatics 23, 184190. doi: 10.1093/bioinformatics/
btl572
Dell'Anna, R., Lazzeri, P., Frisanco, M., Monti, F., Campeggi, F. M., Gottardini, E.,
et al. (2009). Pollen discrimination and classication by Fourier transform
infrared (FT-IR) microspectroscopy and machine learning. Anal. Bioanal.
Chem. 394, 14431452. doi: 10.1007/s00216-009-2794-9
Depciuch, J., Kasprzyk, I., Drzymala, E., and Parlinska-Wojtan, M. (2018).
Identication of birch pollen species using FTIR spectroscopy. Aerobiologia
(Bologna) 34, 525538. doi: 10.1007/s10453-018-9528-4
Diehn, S., Zimmermann, B., Bağcıoğlu, M., Seifert, S., Kohler, A., Ohlson, M., et al.
(2018). Matrix-assisted laser desorption/ionization time-of-ight mass
spectrometry (MALDI-TOF MS) shows adaptation of grass pollen
composition. Sci. Rep. 8, 16591. doi: 10.1038/s41598-018-34800-1
Eilers, P. H. C. (2003). A perfect smoother. Anal. Chem. 75, 36313636. doi:
10.1021/ac034173t
Ellis, D. I., and Goodacre, R. (2006). Metabolic ngerprinting in disease diagnosis:
biomedical applications of infrared and Raman spectroscopy. Analyst 131,
875885. doi: 10.1039/b602376m
Gottardini, E., Rossi, S., Cristofolini, F., and Benedetti, L. (2007). Use of Fourier
transform infrared (FT-IR) spectroscopy as a tool for pollen identication.
Aerobiologia 23, 211219. doi: 10.1007/s10453-007-9065-z
Harrington, P. D. B., Vieira, N. E., Espinoza, J., Nien, J. K., Romero, R., and Yergey,
A. L. (2005). Analysis of variance-principal component analysis: a soft tool for
proteomic discovery. Anal. Chim. Acta 544, 118127. doi: 10.1016/
j.aca.2005.02.042
Hassani, S., Martens, H., Qannari, E. M., Hana, M., Borge, G. I., and Kohler, A.
(2010). Analysis of -omics data: graphical interpretation- and validation tools
in multi-block methods. Chemom. Intell. Lab. Syst. 104, 140153. doi: 10.1016/
j.chemolab.2010.08.008
Hassani, S., Hana, M., Qannari, E., and Kohler, A. (2013). Deation strategies for
multi-block principal component analysis revisited. Chemometrics Intell. Lab.
Syst. 120, 154168. doi: 10.1016/j.chemolab.2012.08.011
Hotelling, H. (1933). Analysis of a complex of statistical variables into principal
components. J. Educ. Psychol. 24, 417441. doi: 10.1037/h0071325
Huang, Y. S., Karashima, T., Yamamoto, M., Ogura, T., and Hamaguchi, H. O.
(2004). Raman spectroscopic signature of life in a living yeast cell. J. Raman
Spectrosc. 35, 525526. doi: 10.1002/jrs.1219
Ivleva, N., Niessner, R., and Panne, U. (2005). Characterization and discrimination
of pollen by Raman microscopy. Anal. Bioanal. Chem. 381, 261267. doi:
10.1007/s00216-004-2942-1
Jansen, J. J., Hoefsloot, H. C. J., Van Der Greef, J., Timmerman, M. E., Westerhuis,
J. A., and Smilde, A. K. (2005). ASCA: analysis of multivariate data obtained
from an experimental design. J. Chemometr. 19, 469481. doi: 10.1002/cem.952
Jardine, P. E., Gosling, W. D., Lomax, B. H., Julier, A. C. M., and Fraser, W. T.
(2019). Chemotaxonomy of domesticated grasses: a pathway to understanding
the origins of agriculture. J. Micropalaentol. 38, 8395. doi: 10.5194/jm-38-83-
2019
Joester, M., Seifert, S., Emmerling, F., and Kneipp, J. (2017). Physiological
inuence of silica on germinating pollen as shown by Raman spectroscopy.
J. Biophotonics 10, 542552. doi: 10.1002/jbio.201600011
Joseph, V., Schulte, F., Rooch, H., Feldmann, I., Dorfel, I., Osterle, W., et al. (2011).
Surface-enhanced Raman scattering with silver nanostructures generated in
situ in a sporopollenin biopolymer matrix. Chem. Commun. 47, 32363238.
doi: 10.1039/c0cc05326k
Julier, A. C. M., Jardine, P. E., Coe, A. L., Gosling, W. D., Lomax, B. H., and Fraser,
W. T. (2016). Chemotaxonomy as a tool for interpreting the cryptic diversity of
Poaceae pollen. Rev. Palaeobot. Palynol. 235, 140147. doi: 10.1016/
j.revpalbo.2016.08.004
Krause, B., Seifert, S., Panne, U., Kneipp, J., and Weidner, S. M. (2012). Matrix-
assisted laser desorption/ionization mass spectrometric investigation of pollen
and their classication by multivariate statistics. Rapid Commun. Mass
Spectrom. 26, 10321038. doi: 10.1002/rcm.6202
Kyu, K. S., Soo, K. M., and Won, S. S. (1987). Surface-enhanced Raman scattering
(SERS) of aromatic amino acids and their glycyl dipeptides in silver sol. J.
Raman Spectrosc. 18, 171175. doi: 10.1002/jrs.1250180305
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178817
Lasch, P., and Naumann, D. (1998). FT-IR microspectroscopic imaging of human
carcinoma thin sections based on pattern recognition techniques. Cell. Mol.
Biol. (Noisy-le-grand) 44, 189202. doi: 10.1063/1.55833
Lauer, F., Diehn, S., Seifert, S., Kneipp, J., Sauerland, V., Barahona, C., et al. (2018).
Multivariate analysis of MALDI imaging mass spectrometry data of mixtures
of single pollen grains. J. Am. Soc. Mass Spectrom. 29, 22372247. doi: 10.1007/
s13361-018-2036-5
Lee, P. C., and Meisel, D. (1982). Adsorption and surface-enhanced Raman of dyes
on silver and gold sols. J. Phys. Chem. 86, 33913395. doi: 10.1021/j100214a025
Li, F.-S., Phyo, P., Jacobowitz, J., Hong, M., and Weng, J.-K. (2019). The molecular
structure of plant sporopollenin. Nat. Plants 5, 4146. doi: 10.1038/s41477-
018-0330-7
Martens, H., and Stark, E. (1991). Extended multiplicative signal correction and
spectral interference subtraction: new preprocessing methods for near infrared
spectroscopy. J. Pharm. Biomed. Anal. 9, 625635. doi: 10.1016/0731-7085(91)
80188-F
Pappas, C. S., Tarantilis, P. A., Harizanis, P. C., and Polissiou, M. G. (2003). New
method for pollen identication by FT-IR spectroscopy. Appl. Spectrosc. 57,
2327. doi: 10.1366/000370203321165160
Pearson, K. (1901). LIII. On lines and planes of closest t to systems of points in
space. London Edinburgh Dublin Philos. Mag. J. Sci. 2, 559572. doi: 10.1080/
14786440109462720
Perisic,N.,Afseth,N.K.,Ofstad,R.,Hassani,S.,andKohler,A.(2013).
Characterising protein, salt and water interactions with combined vibrational
spectroscopic techniques. Food Chem. 138, 679686. doi: 10.1016/
j.foodchem.2012.10.117
Pigorsch, E. (2009). Spectroscopic characterisation of cationic quaternary
ammonium starches. Starch - Stärke 61, 129138. doi: 10.1002/star.200800090
Porra, R. J., Thompson, W. A., and Kriedemann, P. E. (1989). Determination of
accurate extinction coefcients and simultaneous equations for assaying
chlorophylls a and b extracted with four different solvents: verication of the
concentration of chlorophyll standards by atomic absorption spectroscopy.
Biochimica Biophysica Acta (BBA) - Bioenergetics 975, 384394. doi: 10.1016/
S0005-2728(89)80347-0
Pully, V. V., and Otto, C. (2009). The intensity of the 1602 cm1 band in human
cells is related to mitochondrial activity. J. Raman Spectrosc. 40, 473475. doi:
10.1002/jrs.2269
Qian, J., Cutler, J. E., Cole, R. B., and Cai, Y. (2008).MALDI-TOF mass signatures for
differentiation of yeast species,strain grouping and monitoring of morphogenesis
markers. Anal. Bioanal. Chem. 392, 439449. doi: 10.1007/s00216-008-2288-1
Rozema, J., Broekman, R. A., Blokker, P., Meijkamp, B. B., De Bakker, N., Van De
Staaij, J., et al. (2001). UV-B absorbance and UV-B absorbing compounds
(para-coumaric acid) in pollen and sporopollenin: the perspective to track
historic UV-B levels. J. Photochem. Photobiol. B: Biol. 62, 108117. doi:
10.1016/S1011-1344(01)00155-5
Savitzky, A., and Golay, M. J. E. (1964). Smoothing and differentiation of data by
simplied least squares procedures. Anal. Chem. 36, 16271639. doi: 10.1021/
ac60214a047
Schulte, F., Lingott, J., Panne, U., and Kneipp, J. (2008). Chemical characterization
and classication of pollen. Anal. Chem. 80, 95519556. doi: 10.1021/ac801791a
Schulte, F., Mader, J., Kroh, L. W., Panne, U., and Kneipp, J. (2009).
Characterization of pollen carotenoids with in situ and high-performance
thin-layer chromatography supported resonant raman spectroscopy. Anal.
Chem. 81, 84268433. doi: 10.1021/ac901389p
Seifert,S.,Weidner,S.M.,Panne,U.,andKneipp,J.(2015).Taxonomic
relationships of pollens from matrix-assisted laser desorption/ionization
time-of-ight mass spectrometry data using multivariate statistics. Rapid
Commun. Mass Spectrom. 29, 11451154. doi: 10.1002/rcm.7207
Seifert, S., Merk, V., and Kneipp, J. (2016). Identication of aqueous pollen
extracts using surface enhanced Raman scattering (SERS) and pattern
recognition methods. J. Biophotonics 9, 181189. doi: 10.1002/jbio.201500176
Sengupta, A., Laucks, M. L., and Davis, E. J. (2005). Surface-enhanced Raman
spectroscopy of bacteria and pollen. Appl. Spectrosc. 59, 10161023. doi:
10.1366/0003702054615124
Smilde, A. K., Jansen, J. J., Hoefsloot, H. C. J., Lamers, R. J., Van Der Greef, J., and
Timmerman, M. E. (2005). ANOVA-simultaneous component analysis
(ASCA): a new tool for analyzing designed metabolomics data.
Bioinformatics 21, 30433048. doi: 10.1093/bioinformatics/bti476
Solhaug, K. A. (1991). Inuence of photoperiod and temperature on dry matter
production and chlorophyll content in temperate grasses [also incl. net
assimilation rate, NAR, long days, short days]. Norwegian J. Agric. Sci. 5,
365384.
Stewart, S., and Fredericks, P. M. (1999). Surface-enhanced Raman spectroscopy
of amino acids adsorbed on an electrochemically prepared silver surface.
Spectroc. Acta Pt. A-Molec. Biomolec. Spectr. 55, 16411660. doi: 10.1016/
S1386-1425(98)00294-7
Wang, S., Wang, D., Wu, Q., Gao, K., Wang, Z., and Wu, Z. (2015). 3D imaging of
a rice pollen grain using transmission X-ray microscopy. J. Synchrotron Radiat.
22, 10911095. doi: 10.1107/S1600577515009716
Westerhuis, J. A., Kourti, T., and Macgregor, J. F. (1998). Analysis of multiblock
and hierarchical PCA and PLS models. J. Chemometr. 12, 301321. doi:
10.1002/(SICI)1099-128X(199809/10)12:5<301::AID-CEM515>3.0.CO;2-S
Wold, S., Hellberg, S., Lundstedt, T., Sjostrom, M., and Wold, H. (1987). Proc.
symp. on pls model building: theory and application. Frankfurt am Main.
Zimmermann, B., and Kohler, A. (2013). Optimizing Savitzky-Golay parameters
for improving spectral resolution and quantication in infrared spectroscopy.
Appl. Spectrosc. 67, 892902. doi: 10.1366/12-06723
Zimmermann, B., and Kohler, A. (2014). Infrared spectroscopy of pollen identies
plant species and genus as well as environmental conditions. PLoS One 9,
e95417. doi: 10.1371/journal.pone.0095417
Zimmermann, B., Bağcıoğlu, M., Tanstseva, V., Kohler, A., Ohlson, M., and
Fjellheim, S. (2017). A high-throughput FTIR spectroscopy approach to assess
adaptive variation in the chemical composition of pollen. Ecol. Evol. 7, 10839
10849. doi: 10.1002/ece3.3619
Zimmermann, B. (2010). Characterization of pollen by vibrational spectroscopy.
Appl. Spectrosc. 64, 13641373. doi: 10.1366/000370210793561664
Conict of Interest: The authors declare that the research was conducted in the
absence of any commercial or nancial relationships that could be construed as a
potential conict of interest.
Copyright © 2020 Diehn, Zimmermann, Tantseva, Seifert, Bağcıoğlu, Ohlson,
Weidner, Fjellheim, Kohler and Kneipp. This is an open-access article distributed
under the terms of the Creative Commons Attribution License (CC BY). The use,
distribution or reproductionin other forums is permitted, providedthe original author
(s) and the copyright owner(s) are credited and that the original publication in this
journal is cited,in accordance with accepted academic practice. No use,distribution or
reproduction is permitted which does not comply with these terms.
Diehn et al. Multimodal Characterization of Pollen
Frontiers in Plant Science | www.frontiersin.org January 2020 | Volume 10 | Article 178818
... FTIR analysis of pollen is based on the fingerprint-like characteristics of the IR spectrum, containing contributions from all different kinds of biomolecular constituents. More recently, FTIR spectroscopy was shown to allow for a characterization of chemical variation also at the subspecies level, specifically between populations of the same pollen species, and led to conclusions regarding, e.g., the adaptation of plant populations to environmental conditions [22][23][24][25][26][27][28]. Although the majority of FTIR pollen studies were conducted by measurement of bulk pollen samples, containing 1 mg or more of pollen sample per measurement, some studies have used FTIR microspectrometers as well [5,15,18,20,[28][29][30][31][32][33]. ...
... The fact that identification is based here on spectra from individual pollen grains rather than averages from one plant adds another source of variation here, as was recently also discussed when we compared different spectroscopic methods that probe either bulk samples or individual pollen grains and their potential for pollen identification [24]. Nevertheless, the possibility to study pollen spectra in mixtures could in the future open possibilities for FTIR imaging-based identification of mixed grass pollen samples, similar to existing highthroughput and mapping approaches [9,10]. ...
... Therefore, in future work, FTIR microspectroscopy will be combined with other microscopic analytical methods that give single pollen grain spatial resolution and sensitivity, namely Raman microscopy [11,29] and MALDI imaging [9]. Such a multimodal single pollen probing will be a logical continuation of recent experiments combining the complementary chemical information of these methods [18,24]. Table 6 Results of the classification of the spectra of paraffin-embedded pollen corrected using an EMSC model with paraffin constituent spectrum (cf. ...
Article
Fourier-transform infrared (FTIR) spectroscopy enables the chemical characterization and identification of pollen samples, leading to a wide range of applications, such as paleoecology and allergology. This is of particular interest in the identification of grass (Poaceae) species since they have pollen grains of very similar morphology. Unfortunately, the correct identification of FTIR microspectroscopy spectra of single pollen grains is hindered by strong spectral contributions from Mie scattering. Embedding of pollen samples in paraffin helps to retrieve infrared spectra without scattering artifacts. In this study, pollen samples from 10 different populations of five grass species (Anthoxanthum odoratum, Bromus inermis, Hordeum bulbosum, Lolium perenne, and Poa alpina) were embedded in paraffin, and their single grain spectra were obtained by FTIR microspectroscopy. Spectra were subjected to different preprocessing in order to suppress paraffin influence on spectral classification. It is shown that decomposition by non-negative matrix factorization (NMF) and extended multiplicative signal correction (EMSC) that utilizes a paraffin constituent spectrum, respectively, leads to good success rates for the classification of spectra with respect to species by a partial least square discriminant analysis (PLS-DA) model in full cross-validation for several species. PLS-DA, artificial neural network, and random forest classifiers were applied on the EMSC-corrected spectra using an independent validation to assign spectra from unknown populations to the species. Variation within and between species, together with the differences in classification results, is in agreement with the systematics within the Poaceae family. The results illustrate the great potential of FTIR microspectroscopy for automated classification and identification of grass pollen, possibly together with other, complementary methods for single pollen chemical characterization.
... The Raman spectra of the cell walls in tissue sections of different plant species, organs, and histological regions have been discussed, and the contributions by important building blocks of the cell walls to the tissue spectra are becoming better and better understood [9][10][11][12][13][14][15]. The Raman-based observation of small physiological effects on a plant tissue, e.g., of a different genetic background or environmental influences, can prove as a challenge, since different sources of spectral variance are superimposed in a typical data set: In addition to the particular differences under study, a variety of other factors can be named that cause an observed variation of spectra, including the obvious microscopic heterogeneity at both the histological [5] and the cellular level [9], the individual plant-to-plant variation [16], or unwanted contributions to the spectra that differ depending on the sampling specifics, such as variations in background fluorescence, photochemical bleaching, or overall signalto-noise ratio [15]. As a consequence, the identification of 'hierarchical' questions in classification is usually identified, for example, the comparison of spectra from a specific tissue under different conditions [17], or investigations of environmental effects on samples from one population or even one clone [16]. ...
... The Raman-based observation of small physiological effects on a plant tissue, e.g., of a different genetic background or environmental influences, can prove as a challenge, since different sources of spectral variance are superimposed in a typical data set: In addition to the particular differences under study, a variety of other factors can be named that cause an observed variation of spectra, including the obvious microscopic heterogeneity at both the histological [5] and the cellular level [9], the individual plant-to-plant variation [16], or unwanted contributions to the spectra that differ depending on the sampling specifics, such as variations in background fluorescence, photochemical bleaching, or overall signalto-noise ratio [15]. As a consequence, the identification of 'hierarchical' questions in classification is usually identified, for example, the comparison of spectra from a specific tissue under different conditions [17], or investigations of environmental effects on samples from one population or even one clone [16]. Furthermore, different data pre-treatment strategies are in place and have been applied, ranging from smoothing of original spectra or their derivatives using the Savitzky-Golay algorithm [18], smoothing using asymmetric least squares (AsLS) [2,[19][20][21][22][23], as well as sorting and data extraction based on spectral patterns [15]. ...
... It should be emphasized that in this PCA neither the spatial information, allowing conclusions about the different histological substructures between and within maps, nor the variation caused by individual variability has been considered so far. As shown in previous spectroscopic studies on physiological differences revealing in the spectra of other plant tissues, in particular pollen [16,55], where we have assessed the influence of different sources of variance, the spectral differences caused by different individuals from the same population can dominate the variance, compared to, e.g., the effects of altered growth conditions. Fig. 4E and 4F show the results of a PCA of a subset of the complete leaf data, comprising only the spectra from one map of leaf tissue grown in the presence of additional silicic acid (Fig. 4E, black marking) and from one map of a plant grown in the absence of silicic acid (Fig. 4E, blue markings). ...
Article
Identifying and characterizing the biochemical variation in plant tissues is an important task in many research fields. Small spectral differences of the plant cell wall that are caused by genetic or environmental influences may be superimposed by individual variation as well as by a microscopic heterogeneity in molecular composition and structure of different histological substructures. A set of 56 samples from Cucumis sativus (cucumber) plants, comprising a total of ∼168000 spectra from tissue sections of leaf, stem, and roots was investigated by Raman microspectroscopic mapping excited at 532 nm. A multivariate analysis was carried out in order to assess the variation of the spectra with respect to origin of the tissue, the histological (cell wall) substructures, and the possibility to discriminate the spectra obtained from different individuals that had been subjected to two different conditions during growth. Combining the results of principal component analysis (PCA) based classification with the original spatial information in the maps of 23 sections of leaf xylem, variation in cell wall composition is found for four different individuals that also includes a discrimination of tissue grown in the presence and absence of additional silicic acid in the irrigation water of the plants. The spectral data point to differences in a contribution by carotenoids, as well as by hydroxycinnamic acids to the spectra. The results give new insight into the chemical heterogeneity of plant tissues and may be useful for elucidating biochemical processes associated with biomineralization by vibrational spectroscopy.
... Para el análisis del conjunto de datos espectrales de corteza y mesófilo sometidos al preprocesamiento de datos, se seleccionó la región espectral de 1800-800 cm -1 (Diehn et al., 2020a). Se aplicó el PCA para estimar la similitud química entre los conjuntos de datos y entre las especies en cada conjunto de datos. ...
... Sin embargo, se observaron algunos ligeros desplazamientos en el rango de frecuencia de las bandas, los cuales se consideran normales (Delgado et al., 2014). En este estudio se realizó el análisis químico por espectroscopía FTIR de la corteza y del mesófilo de las hojas, donde las huellas químicas en la región espectral de 1800-800 cm -1 fueron especie-específicas (Zimmermann et al., 2017;Diehn et al., 2020a). ...
Thesis
El objetivo de este estudio fue determinar la composición química y el perfil genético de 16 especies, pertenecientes a cinco géneros, de la familia Asphodelaceae. La composición química se determinó empleando ensayos colorimétricos, cromatografía de capa fina (TLC) y espectroscopía infrarroja con transformada de Fourier (FTIR); por otra parte, el perfil genético fue determinado utilizando el marcador molecular de DNA de repetición entre secuencias simples (ISSR). Dicho perfil genético, permitió agrupar a las especies estudiadas a nivel de género. La determinación de la capacidad antioxidante y del contenido de carbohidratos, proteínas, compuestos fenólicos y flavonoides indicó variación entre las especies estudiadas, los niveles de estas determinaciones no fueron consistentes por género (Aloe, x Alworthia, Bulbine, Gasteria y Haworthia). La TLC permitió identificar carbohidratos como glucosa, fructosa, sacarosa y fructooligosacáridos para algunas especies, así como identificar la presencia de compuestos fenólicos, donde predominaron los derivados de flavonoles y flavonas. En el análisis por espectroscopía FTIR, que se realizó en la corteza y en el mesófilo de once especies de la familia Asphodelaceae, se identificaron carbohidratos, lípidos, proteínas y compuestos fenólicos para la corteza y carbohidratos y lípidos para el mesófilo. Las huellas químicas de las cortezas, generadas por espectroscopía FTIR, fueron especie-especificas, lo que permitió ordenarlas de acuerdo con la taxonomía para la familia Asphodelaceae. Los grupos generados por el PCA de los espectros de corteza y del perfil genético tuvieron una correlación de 0.522, sugiriendo que la espectroscopía FTIR tiene un importante potencial taxonómico y que el perfil químico depende del genotipo. Las especies con mayor similitud química con Aloe vera, como A. vera var. chinensis, x Alworthia, B. frutescens y H. cymbiformis, que presentaron compuestos de interés como fructooligosacáridos y compuestos fenólicos, podrían usarse en investigaciones futuras para determinar sus propiedades biológicas y/o medicinales.
... A small subset of studies dealt with different enhancement techniques, such as surface-enhanced Raman spectroscopy [18][19][20] or resonance-enhanced Raman spectroscopy [21]. The majority of studies focused on the taxonomic classification and their relationships [22][23][24][25][26], and there were few studies on the differentiation due to, e.g., growth and environmental conditions [27][28][29] or nutritional parameters [30]. Recently, efforts towards a rapid and automatic identification of pollen were made based on the reduction of the spectral dataset into specific band regions [31], or on the utilization of a high-throughput IR system [32]. ...
Article
Full-text available
The investigation of the biochemical composition of pollen grains is of the utmost interest for several environmental aspects, such as their allergenic potential and their changes in growth conditions due to climatic factors. In order to fully understand the composition of pollen grains, not only is an in-depth analysis of their molecular components necessary but also spatial information of, e.g., the thickness of the outer shell, should be recorded. However, there is a lack of studies using molecular imaging methods for a spatially resolved biochemical composition on a single-grain level. In this study, Raman spectroscopy was implemented as an analytical tool to investigate birch pollen by imaging single pollen grains and analyzing their spectral profiles. The imaging modality allowed us to reveal the layered structure of pollen grains based on the biochemical information of the recorded Raman spectra. Seven different birch pollen species collected at two different locations in Germany were investigated and compared. Using chemometric algorithms such as hierarchical cluster analysis and multiple-curve resolution, several components of the grain wall, such as sporopollenin, as well as the inner core presenting high starch concentrations, were identified and quantified. Differences in the concentrations of, e.g., sporopollenin, lipids and proteins in the pollen species at the two different collection sites were found, and are discussed in connection with germination and other growth processes.
... Ideally, their results are also analyzed simultaneously, or at least in combination, to maximize information gains about lignin content and composition in situ. A recent example (albeit not specifically focusing on lignin) combines the information from FTIR and Raman spectroscopies (as well as surface enhanced Raman scattering and matrix assisted laser desorption/ionization mass spectrometry) to characterize the pollens of Poa alpine grass grown under different conditions (Diehn et al., 2020). Similar approaches would benefit lignin imaging in situ, as the different techniques highlight different aspects of lignin chemistry and structure with different sensitivities, providing valuable complementary information. ...
Chapter
Spectral imaging technologies simultaneously record spectral and spatial information about plant tissues in a noninvasive way. Differences in techniques result in different selection rules and spatial resolutions. This article introduces the basic principles of Raman, Fourier-transform Infrared (FTIR) and autofluorescence imaging and finally compares their strength and drawbacks. The methods result in spectral datasets as bases for image generation. Spectral preprocessing together with univariate and multivariate data analysis approaches are essential for informative lignin imaging and analysis and therefore also briefly illustrated and discussed. Examples of imaging lignin and other aromatic components in a broad range of plant tissues show the potential as well as the limitations of microspectroscopic imaging.
... Extensive research shows the ability of FTIR to classify pollen species and their quality and characterize the grains according to genetic background and environmental conditions. [117][118][119][120][121] Raman is more suitable to perform in situ studies of fresh plant material because water has only weak Raman scattering properties. 122 However, the excitation wavelength used may interfere and induce fluorescence. ...
Thesis
Die Ablagerung von Siliziumdioxid ist ein verbreitetes Phänomen, das mit der Toleranz von Pflanzen gegenüber Belastungen korreliert. Die Pflanzen akkumulieren das amorphe Siliziumdioxid in mikroskopischen Partikeln, den Phytolithen, jedoch ist der exakte Mechanismus nicht vollständig aufgeklärt. Um ein besseres Verständnis über die Ablagerung von Siliziumdioxid zu erlangen, wurden verschiedene spektroskopische Techniken an Sorghumblättern und molekularen Modellen angewandt. Festkörper Kernspinresonanz und thermogravimetrische Analysen zeigen, dass die Siliziumdioxidstruktur von der Phytolithe-Extraktion abhängt. Basierend auf Raman- und IR-Daten einzelner Phytolithe lassen sich die Änderungen dieser Strukturen ermitteln. Das deutet auf unterschiedliche biologische Prozesse der Ablagerung des Siliciumdioxids hin. Die Pflanzengewebe in denen Siliciumdioxid abgelagert ist, wurden mit einem multimodalen Ansatz charakterisiert, welcher Fluoreszenz-, Hellfeld- und Rasterelektronenmikroskopie beinhaltet. Die chemische Zusammensetzung der Pflanzengewebe wurden mit Raman- und FTIR-Mikrospektroskopie kartiert. Ein neuartiger Ansatz zur Untersuchung von Pflanzengeweben wurde verwendet, basierend auf der optischen Nahfeldmikroskopie im mittleren IR-Bereich. Dieser ermöglicht eine kombinierte Analyse von mechanischen Materialeigenschaften sowie der chemischen Zusammensetzung und Struktur. Um die Rolle der organischen Matrix zu verstehen, wurden Modellverbindungen betrachtet, die die Ablagerung von Kieselsäure in den Pflanzen induzieren können. In-vitro-Reaktionen konnten eine gleichzeitige Präzipitation von Lignin und Siliciumdioxid sowie eine Polymerisation zusammen mit Peptiden simulieren. Die Ergebnisse lassen starke Wechselwirkungen zwischen diesen Verbindungen vermuten. Neben einem besseren Verständnis verschiedener Aspekte der Silifizierung von Pflanzen werden in dieser Arbeit neue Methoden zur Charakterisierung von Pflanzenproben vorgeschlagen.
... After averaging of technical replicates, FTIR and FT-Raman spectral sets can be analyzed as a multiblock data with a sample-to-sample correspondence between the data blocks. Consensus principal component analysis (CPCA) is a frequently used multiblock data analysis method since it allows assessment of the covariance patterns using more than one block of data [77,78]. CPCA provides global scores that describe the consensus of all data blocks involved in the CPCA. ...
Article
Full-text available
Oleaginous filamentous fungi can accumulate large amount of cellular lipids and biopolymers and pigments and potentially serve as a major source of biochemicals for food, feed, chemical, pharmaceutical, and transport industries. We assessed suitability of Fourier transform (FT) Raman spectroscopy for screening and process monitoring of filamentous fungi in biotechnology. Six Mucoromycota strains were cultivated in microbioreactors under six growth conditions (three phosphate concentrations in the presence and absence of calcium). FT-Raman and FT-infrared (FTIR) spectroscopic data was assessed in respect to reference analyses of lipids, phosphorus, and carotenoids by using principal component analysis (PCA), multiblock or consensus PCA, partial least square regression (PLSR), and analysis of spectral variation due to different design factors by an ANOVA model. All main chemical biomass constituents were detected by FT-Raman spectroscopy, including lipids, proteins, cell wall carbohydrates, and polyphosphates, and carotenoids. FT-Raman spectra clearly show the effect of growth conditions on fungal biomass. PLSR models with high coefficients of determination (0.83–0.94) and low error (approximately 8%) for quantitative determination of total lipids, phosphates, and carotenoids were established. FT-Raman spectroscopy showed great potential for chemical analysis of biomass of oleaginous filamentous fungi. The study demonstrates that FT-Raman and FTIR spectroscopies provide complementary information on main fungal biomass constituents.
... The wavenumbers where visible changes are observed can be assigned to nucleic acids adenine and guanine (~1360 cm −1 ) [45], amide bands of proteins (amide I, II and III corresponding to C = O stretch ~1650 cm −1 , N-H and C-N deformation ~1550 cm −1 and 1300 cm −1 ), and deformation made of C-H2 groups of aliphatic carbon chains [46], sporopollenin or lipids (1440 cm −1 ) [47]. The pollen wall has a complex chemical composition, which is important to the pollen wall's structural maintenance and protection of reproductive nuclei. ...
Article
Full-text available
The knowledge of pollen sensitivity and tolerance to stress factors such as air pollution is important for forest sustainability, ensuring the most efficient production with the highest benefits and lowest resource losses. This study intended to evaluate the influence of common air pollutants in four forest trees species, Betula pendula Roth, Corylus avellana L., Acer negundo L. and Quercus robur L., through a comparative analysis at the same experimental conditions. We aimed to investigate the effect that may occur in pollen fertility, protein content, oxidative stress and wall composition after exposure in vitro to ozone and nitrogen dioxide at concentration levels for vegetation protection in Europe. Our results suggest changes in pollen viability, protein content and differential sensitivity related to ROS synthesis, NADPH oxidase activity, as well as in wall composition. The results indicate that NO2 exposure affected more the pollen species studied mostly at the highest concentration exposure. As for ozone, there were less significant differences between samples; however, a different behavior occurs in O3 expositions, where the most influence happens at the legal limit for vegetation protection in Europe. Our study showed that significant pollen functions could be compromised even at common air pollutant’s concentrations.
... The spectacular inertness of sporopollenin toward the most disparate analytical techniques made the search for its molecular structure hard to obtain, and at the same time, very desirable given the multitude of promising applications in the fields of material engineering and nanotechnology. Progress has also been made in the analysis of lipids, proteins, and the mechanisms of accumulation of flavonoid glycosides on the surface of pollen grains [90][91][92][93], as well as the metabolic processes underlying pollen germination and pollen tube elongation [94]. However, given the complexity of these chemical signals, their physiological function is not yet fully understood [95]. ...
Article
Full-text available
Omics approaches for investigating biological systems were introduced in the mid‐1990s and quickly consolidated to become a fundamental pillar of modern biology. The idea of measuring the whole complement of genes, transcripts, proteins, and metabolites has since become widespread and routinely adopted in the pursuit of an infinity of scientific questions. Incremental improvements over technical aspects such as sampling, sensitivity, cost, and throughput pushed even further the boundaries of what these techniques can achieve. In this context, single‐cell genomics and transcriptomics quickly became a well‐established tool to answer fundamental questions challenging to assess at a whole tissue level. Following a similar trend as the original development of these techniques, proteomics alternatives for single‐cell exploration have become more accessible and reliable, whilst metabolomics lag behind the rest. This review summarizes state‐of‐the‐art technologies for spatially resolved metabolomics analysis, as well as the challenges hindering the achievement of sensu stricto metabolome coverage at the single‐cell level. Furthermore, we discuss several essential contributions to understanding plant single‐cell metabolism, finishing with our opinion on near‐future developments and relevant scientific questions that will hopefully be tackled by incorporating these new exciting technologies.
Article
The plant cell wall is a complex composite material made of polysaccharides, polyphenols, proteins, and minerals. In this work, a multimodal imaging approach was taken, using Raman and Fourier‐transform infrared (FTIR) microspectroscopy along with fluorescence imaging, scanning electron microscopy (SEM), and elemental mapping by energy dispersive X‐ray spectroscopy (EDX). We characterized the chemical composition of sorghum leaf cross‐sections extracted from fresh tissue as well as after paraffin embedding. The complementary vibrational information of Raman and FTIR spectra related a silica deposition to a specific organic composition in the epidermis, specifically with respect to lignin. Moreover, the data enable in situ correlation of autofluorescence with a specific lignin structure. Our results showed that a lignin 5‐5’ linkages that produce biphenyl structures are important determinants of the cell wall fluorescence properties. The reported multimodal approach will help to clarify the process of biosilica formation and related questions regarding cell wall biochemistry.
Article
Full-text available
The grass family (Poaceae) is one of the most economically important plant groups in the world today. In particular many major food crops, including rice, wheat, maize, rye, barley, oats and millet, are grasses that were domesticated from wild progenitors during the Holocene. Archaeological evidence has provided key information on domestication pathways of different grass lineages through time and space. However, the most abundant empirical archive of floral change – the pollen record – has been underused for reconstructing grass domestication patterns because of the challenges of classifying grass pollen grains based on their morphology alone. Here, we test the potential of a novel approach for pollen classification based on the chemical signature of the pollen grains measured using Fourier transform infrared (FTIR) microspectroscopy. We use a dataset of eight domesticated and wild grass species, classified using k-nearest neighbour classification coupled with leave-one-out cross validation. We demonstrate a 95 % classification success rate on training data and an 82 % classification success rate on validation data. This result shows that FTIR spectroscopy can provide enhanced taxonomic resolution enabling species level assignment from pollen. This will enable the full testing of the timing and drivers of domestication and agriculture through the Holocene.
Article
Full-text available
Sporopollenin is a ubiquitous and extremely chemically inert biopolymer that constitutes the outer wall of all land-plant spores and pollen grains¹. Sporopollenin protects the vulnerable plant gametes against a wide range of environmental assaults, and is considered a prerequisite for the migration of early plants onto land². Despite its importance, the chemical structure of plant sporopollenin has remained elusive¹. Using a newly developed thioacidolysis degradative method together with state-of-the-art solid-state NMR techniques, we determined the detailed molecular structure of pine sporopollenin. We show that pine sporopollenin is primarily composed of aliphatic-polyketide-derived polyvinyl alcohol units and 7-O-p-coumaroylated C16 aliphatic units, crosslinked through a distinctive dioxane moiety featuring an acetal. Naringenin was also identified as a minor component of pine sporopollenin. This discovery answers the long-standing question about the chemical make-up of plant sporopollenin, laying the foundation for future investigations of sporopollenin biosynthesis and for the design of new biomimetic polymers with desirable inert properties. © 2018, The Author(s), under exclusive licence to Springer Nature Limited.
Article
Full-text available
MALDI time-of-flight mass spectrometry (MALDI-TOF MS) has become a widely used tool for the classification of biological samples. The complex chemical composition of pollen grains leads to highly specific, fingerprint-like mass spectra, with respect to the pollen species. Beyond the species-specific composition, the variances in pollen chemistry can be hierarchically structured, including the level of different populations, of environmental conditions or different genotypes. We demonstrate here the sensitivity of MALDI-TOF MS regarding the adaption of the chemical composition of three Poaceae (grass) pollen for different populations of parent plants by analyzing the mass spectra with partial least squares discriminant analysis (PLS-DA) and principal component analysis (PCA). Thereby, variances in species, population and specific growth conditions of the plants were observed simultaneously. In particular, the chemical pattern revealed by the MALDI spectra enabled discrimination of the different populations of one species. Specifically, the role of environmental changes and their effect on the pollen chemistry of three different grass species is discussed. Analysis of the group formation within the respective populations showed a varying influence of plant genotype on the classification, depending on the species, and permits conclusions regarding the respective rigidity or plasticity towards environmental changes.
Article
Full-text available
In this study, the morphology and chemical composition of pollen grains of six birch species (Betula utilis Doorenbos, B. dahurica, B. maximowicziana, B. pendula, B. pubescens and B. humilis) were examined to verify which of these features allow distinguishing them in a more unambiguous way. For this purpose, scanning electron microscopy and light microscopy, as well as Fourier transform infrared (FTIR) spectroscopy and curve-fitting analysis of amide I profile, were performed. The microscopy images show that the pollen grains of B. pubescens, B. pendula and B. humilis are similar in diameter and significantly smaller than those of others species, with the largest diameter observed for B. utilis Doorenbos. However, the results obtained from FTIR spectroscopy indicate that the chemical compositions of B. pubescens and B. pendula are similar, but B. humilis is outlaying. Summarizing, it is not possible to unambiguously state, which feature or which technique is the best for differentiating between the six chosen birch species. However, the study showed that both techniques have potential for identification of birch pollen species.
Article
Full-text available
The two factors defining male reproductive success in plants are pollen quantity and quality, but our knowledge about the importance of pollen quality is limited due to methodological constraints. Pollen quality in terms of chemical composition may be either genetically fixed for high performance independent of environmental conditions, or it may be plastic to maximize reproductive output under different environmental conditions. In this study, we validated a new approach for studying the role of chemical composition of pollen in adaptation to local climate. The approach is based on high-throughput Fourier infrared (FTIR) characterization and biochemical interpretation of pollen chemical composition in response to environmental conditions. The study covered three grass species, Poa alpina, Anthoxanthum odoratum, and Festuca ovina. For each species, plants were grown from seeds of three populations with wide geographic and climate variation. Each individual plant was divided into four genetically identical clones which were grown in different controlled environments (high and low levels of temperature and nutrients). In total, 389 samples were measured using a high-throughput FTIR spectrometer. The biochemical fingerprints of pollen were species and population specific, and plastic in response to different environmental conditions. The response was most pronounced for temperature, influencing the levels of proteins, lipids, and carbohydrates in pollen of all species. Furthermore, there is considerable variation in plasticity of the chemical composition of pollen among species and populations. The use of high-throughput FTIR spectroscopy provides fast, cheap, and simple assessment of the chemical composition of pollen. In combination with controlled-condition growth experiments and multivariate analyses, FTIR spectroscopy opens up for studies of the adaptive role of pollen that until now has been difficult with available methodology. The approach can easily be extended to other species and environmental conditions and has the potential to significantly increase our understanding of plant male function.
Article
Full-text available
The uniform morphology of different species of Poaceae (grass) pollen means that identification to below family level using light microscopy is extremely challenging. Poor taxonomic resolution reduces recoverable information from the grass pollen record, for example, species diversity and environmental preferences cannot be extracted. Recent research suggests Fourier Transform Infra-red Spectroscopy (FTIR) can be used to identify pollen grains based on their chemical composition. Here, we present a study of 12 species from 8 subfamilies of Poaceae, selected from across the phylogeny but from a relatively constrained geographical area (tropical West Africa) to assess the feasibility of using this chemical method for identification within the Poaceae family. We assess several spectral processing methods and use K-nearest neighbour (k-nn) analyses, with a leave-one-out cross-validation, to generate identification success rates at different taxonomic levels. We demonstrate we can identify grass pollen grains to subfamily level with an 80% success rate. Our success in identifying Poaceae to subfamily level using FTIR provides an opportunity to generate high taxonomic resolution datasets in research areas such as palaeoecology, forensics, and melissopalynology quickly and at a relatively low cost.
Article
Full-text available
Background Analysis of pollen grains reveals valuable information on biology, ecology, forensics, climate change, insect migration, food sources and aeroallergens. Vibrational (infrared and Raman) spectroscopies offer chemical characterization of pollen via identifiable spectral features without any sample pretreatment. We have compared the level of chemical information that can be obtained by different multiscale vibrational spectroscopic techniques. Methodology Pollen from 15 different species of Pinales (conifers) were measured by seven infrared and Raman methodologies. In order to obtain infrared spectra, both reflectance and transmission measurements were performed on ground and intact pollen grains (bulk measurements), in addition, infrared spectra were obtained by microspectroscopy of multigrain and single pollen grain measurements. For Raman microspectroscopy measurements, spectra were obtained from the same poll
Article
Mixtures of pollen grains of three different species (Corylus avellana, Alnus cordata, and Pinus sylvestris) were investigated by matrix-assisted laser desorption/ionization time-of-flight imaging mass spectrometry (MALDI-TOF imaging MS). The amount of pollen grains was reduced stepwise from > 10 to single pollen grains. For sample pretreatment, we modified a previously applied approach, where any additional extraction steps were omitted. Our results show that characteristic pollen MALDI mass spectra can be obtained from a single pollen grain, which is the prerequisite for a reliable pollen classification in practical applications. MALDI imaging of laterally resolved pollen grains provides additional information by reducing the complexity of the MS spectra of mixtures, where frequently peak discrimination is observed. Combined with multivariate statistical analyses, such as principal component analysis (PCA), our approach offers the chance for a fast and reliable identification of individual pollen grains by mass spectrometry. Open image in new window Graphical Abstract ᅟ
Article
Fourier transform infrared (FTIR) spectroscopy enables chemical analysis of pollen samples for plant phenotyping to study plant-environment interactions, such as influence of climate change or pathogens. However, current approach, such as microspectroscopy and Attenuated Total Reflection spectroscopy, does not allow for high-throughput protocols. The paper at hand suggests a new spectroscopic method for high-throughput characterization of pollen. 2. Samples were measured as thin films of pollen fragments using a Bruker FTIR spectrometer with a high-throughput eXTension (HTS-XT) unit employing 384-well plates. In total, 146 pollen samples, belonging to 31 different pollen species of Fagaceae and Betulaceae and collected during three consecutive years (2012-2014) at locations in Croatia, Germany and Norway, were analyzed. Critical steps in the sample preparation and measurement, such as variabilities between technical replicates, between microplates and between spectrometers, were studied. 3. Measurement variations due to sample preparation, microplate holders and instrumentation were low, and thus allowed differentiation of samples with respect to phylogeny and biogeography. The spectral variability for a