TYPE
Original Research
02 September 2022
10.3389/fpls.2022.1001023
PUBLISHED
DOI
OPEN ACCESS
EDITED BY
Milen I. Georgiev,
Bulgarian Academy of Sciences, Bulgaria
REVIEWED BY
Shakti Mehrotra,
Institute of Engineering and Technology,
Lucknow, India
Aleksandra Krolicka,
University of Gdansk,
Poland
Using machine learning to link
the influence of transferred
Agrobacterium rhizogenes genes
to the hormone profile and
morphological traits in Centella
asiatica hairy roots
*CORRESPONDENCE
Javier Palazon
javierpalazon@ub.edu
Diego Hidalgo-Martinez
dhidalgo@ub.edu
SPECIALTY SECTION
This article was submitted to
Plant Biotechnology,
a section of the journal
Frontiers in Plant Science
RECEIVED 22
July 2022
August 2022
PUBLISHED 02 September 2022
ACCEPTED 17
Miguel Angel Alcalde 1, Maren Müller 2, Sergi Munné-Bosch 2,
Mariana Landín 3, Pedro Pablo Gallego 4, Mercedes Bonfill 1,
Javier Palazon 1* and Diego Hidalgo-Martinez 1,5*
1
Department of Biology, Healthcare and the Environment, Faculty of Pharmacy and Food Sciences,
University of Barcelona, Barcelona, Spain, 2 Department of Evolutionary Biology, Ecology and
Environmental Sciences, Faculty of Biology, University of Barcelona, Barcelona, Spain, 3 Department
of Pharmacology, Pharmacy and Pharmaceutical Technology, Group I+D Farma (GI-1645), Faculty
of Pharmacy, University of Santiago, Santiago de Compostela, Spain, 4 Agrobiotech for Health,
Department of Plant Biology and Soil Science, Faculty of Biology, University of Vigo, Vigo, Spain,
5
Department of Plant and Microbial Biology, University of California, Berkeley, Berkeley, CA,
United States
CITATION
Alcalde MA, Müller M, Munné-Bosch S,
Landín M, Gallego PP, Bonfill M,
Palazon J and Hidalgo-Martinez D (2022)
Using machine learning to link the
influence of transferred Agrobacterium
rhizogenes genes to the hormone profile
and morphological traits in Centella asiatica
hairy roots.
Front. Plant Sci. 13:1001023.
doi: 10.3389/fpls.2022.1001023
COPYRIGHT
© 2022 Alcalde, Müller, Munné-Bosch,
Landín, Gallego, Bonfill, Palazon and
Hidalgo-Martinez. This is an open-access
article distributed under the terms of the
Creative Commons Attribution License (CC
BY). The use, distribution or reproduction in
other forums is permitted, provided the
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.
Frontiers in Plant Science
Hairy roots are made after the integration of a small set of genes from
Agrobacterium rhizogenes in the plant genome. Little is known about how this
small set is linked to their hormone profile, which determines development,
morphology, and levels of secondary metabolite production. We used C.
asiatica hairy root line cultures to determine the putative links between the rol
and aux gene expressions with morphological traits, a hormone profile, and
centelloside production. The results obtained after 14 and 28days of culture
were processed via multivariate analysis and machine-learning processes such
as random forest, supported vector machines, linear discriminant analysis, and
neural networks. This allowed us to obtain models capable of discriminating
highly productive root lines from their levels of genetic expression (rol and aux
genes) or from their hormone profile. In total, 12 hormones were evaluated,
resulting in 10 being satisfactorily detected. Within this set of hormones,
abscisic acid (ABA) and cytokinin isopentenyl adenosine (IPA) were found to
be critical in defining the morphological traits and centelloside content. The
results showed that IPA brings more benefits to the biotechnological platform.
Additionally, we determined the degree of influence of each of the evaluated
genes on the individual hormone profile, finding that aux1 has a significant
influence on the IPA profile, while the rol genes are closely linked to the ABA
profile. Finally, we effectively verified the gene influence on these two specific
hormones through feeding experiments that aimed to reverse the effect on
root morphology and centelloside content.
01
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
KEYWORDS
Agrobacterium rhizogenes, plant hormones, hairy root cultures, Centella asiatica,
centellosides, machine learning, random forest
Introduction
genome. rolA is found on all Ri plasmids and encodes a small
protein with a basic isoelectric point whose expression showed a
dramatic reduction in several classes of hormones (Ozyigit et al.,
2013). rolB may play a critical role in the early stages of hairy-root
induction and is the most powerful inducer of secondary
metabolism (Dilshad et al., 2021). rolC is considered the most
conserved of all rol genes and has a minor impact on root
formation (Makhzoum et al., 2013). T-DNA also contains genes
that encode enzymes that direct the production of opines, which
are synthesized and excreted by transformed cells and consumed
as a source of carbon and nitrogen by A. rhizogenes (Matveeva and
Otten, 2021). The second region of T-DNA is TR-DNA which
contains genes related to auxin biosynthesis, known as aux1 and
aux2. Both regions can be transferred to the nuclear genome of
infected plant cells independently (Nemoto et al., 2009).
Most studies on the expression of rol and aux genes in hairy
root cultures have focused on demonstrating their effect on
specialized metabolism for increasing the production of
phytochemicals, such as centelloside production in C. asiatica
hairy root cultures (Dilshad et al., 2021), anthraquinones in rolAtransgenic tissues of Rubia cordifolia calli (Nemoto et al., 2009), or
nicotine in tobacco hairy roots (Palazón et al., 1997). The rolB
transformation was shown to induce resveratrol production in
Vitis amurensis cells (Kiselev et al., 2007), and rolC gene expression
was shown to be highly efficient for increasing the production of
morphinan, tropane alkaloids (Cardillo et al., 2013; Hashemi and
Naghavi, 2016), anthraquinones (Bulgakov et al., 2002).
On the contrary, the relationship between rol and aux genes
expression and the hormonal profile, which is a determinant in
root development and morphology (Wahby et al., 2012), has
received little attention. As a result, we focus on analyzing the
relationship between: (1) expression levels of different rol and aux
genes, (2) morphological traits, (3) production of triterpene
saponins and (4) hormonal profiling of various hairy root lines of
C. asiatica. The combined analysis of these variables enabled us to
generate machine learning models that allow for the
discrimination of producing lines or lines with improved traits,
either by the level of gene expression or hormonal profile. The
extent to which A. rhizogenes genes influenced each of the
hormones measured has been determined.
Plant specialized metabolism is the source of a plethora of
bioactive compounds, some of which are uncommon and with
important pharmacological activities (Rai et al., 2017). The
overuse of medicinal plants and the fact that some of their
bioactive compounds are only found in trace amounts in plant
tissues has prompted the search for alternative sources of these
compounds (Chen et al., 2016).
This is the case with Centella asiatica (L.) Urban, which has
been used in traditional medicine to treat several chronic diseases
since ancient times (Prasad et al., 2019). C. asiatica extracts have
antidepressant,
antiepileptic,
antidiabetic,
anxiolytic,
neuroprotective, antioxidant, antiulcer, antitumor, antiinflammatory, and healing properties (Gallego et al., 2014; Sun
et al., 2020; Arribas-López et al., 2022), about 139 metabolites has
been isolated from this plant most of them extracted from leaves
and roots (Kunjumon et al., 2022). C. asiatica is well known for
the accumulation of triterpenoid centelloside, such as
madecasosside and asiaticoside, as well as its relevant aglycones
madecassic and asiatic acid (Joshi and Chaturvedi, 2013), which
contribute to its clinical efficacy (Prakash et al., 2017). As a result,
host plants have been over-exploited, and excessive uprooting has
put C. asiatica in danger of extinction (Mangas et al., 2008). The
chemical synthesis of centellosides is either impossible or
economically unviable. Research has focused on the potential
offered by plant cell culture technology, also known as Plant
Biofactories, for efficient specialized metabolite production, such
as centellosides (Dhillon et al., 2017).
One of these technologies is the generation of hairy roots,
which has been used in several biotechnological approaches for
phytochemical production (Donini and Marusic, 2018; Roy,
2020). Hairy root cultures are induced in most dicotyledonous
plants by incorporating a segment of Agrobacterium rhizogenes
DNA (T-DNA) from the plasmid Ri-DNA into the plant cell
genome, where the expression of the genes carried out by the
T-DNA promotes rooting at the site of infection. Hairy roots can
grow rapidly even in the absence of exogenous growth regulators,
which is why they are widely used as a transgenic tool to produce
specialized metabolites, therapeutic proteins, etc. (GutierrezValdes et al., 2020).
The plasmid Ri-DNA of A. rhizogenes strain A4 contains two
regions: TL-DNA and TR-DNA. The first contains four rol genes
(rooting locus): A, B, C, and D, which improve plant cell
susceptibility to auxins and cytokinins and are responsible for the
formation of these roots (Supplementary Figure 1; Mauro et al.,
2017). However, little is known about the molecular changes
induced in plant cells by the expression of rol genes into the plant
Frontiers in Plant Science
Materials and methods
Establishment of hairy root culture
The A. rhizogenes A4 strain was used in transformation
experiments. Bacteria were grown for 48 h (OD 600 = 0.5–0.6)
02
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
in liquid YEB (Yeast Extract Beef ) medium at 28°C on a
rotary shaker at 130 rpm. Explants for co-cultivation and
hairy root induction were leaf segments from a in-vitro
2-month-old seedling of C. asiatica. The explants were cut
into 1.5–2 cm2 disks with the tip of a scalpel containing a
colony of A. rhizogenes, cultured at 25°C. All excised explants
were then co-cultured in solid MS hormone-free media
enriched with 3% sucrose and pH = 5.8. The explants were
transferred to a fresh solid MS medium containing 500 mg/l
cefotaxime after 48 h of cocultivation in the dark at 28°C. The
emerging hairy roots were excised and transferred to a fresh
solid MS medium containing 500 mg/l cefotaxime, where they
were grown in darkness at 25°C on a rotary shaker. To
eliminate the bacteria, this step was repeated every 2 weeks
for about 2 months.
agarose gels in TAE buffer (1X), and pictures were taken using
a Gel Logic 100 camera (KODAK). The bands were quantified
using the Kodak Gel Logic 100 Digital Imaging System
software (KODAK).
Evaluation of some morphology
parameters of hairy root lines
An inoculum of 10 mg fresh weight (FW) from each hairy
root line was placed in plates with MS medium solid, and the
cultures were maintained for two subcultures every 2 weeks at
25°C in dark conditions, as we had done in previous studies
(Alcalde et al., 2022), before evaluating the growth parameters
considered using three replicas of each line. The branching rate
was defined as the number of lateral roots per cm of initial stem
root (number of lateral roots/cm); the growth rate as the average
length of the lateral roots (mm/day); and the biomass productivity
as the final FW minus initial FW divided by the number of
growing days (mg/day).
Semiquantitative RT-PCR detection and
expression of transgene integrations
Semiquantitative RT-PCR was used to detect the
integration and expression of A. rhizogenes T-DNA genes
(rolA, rolB, rolC, and aux1) at the transcript level in the
studied transgenic clones, this analysis was previously perform
in C. asiatica by Mangas et al. (2008). PureLink RNA Mini Kit
(Invitrogen) was used to isolate RNA from 200 mg of fresh
hairy roots lines according to the manufacturer’s instructions.
The amount and quality of each RNA sample were determined
using the NanoDrop 2000 Spectrophotometer (Thermo
Scientific). The integrity of the RNA was assessed using
agarose gel electrophoresis. The total RNA at a fixed
concentration (1.5 μg of RNA) was used as the template for the
DNAse treatment. For this purpose, the required sample
volume was calculated, taking into account the volume of
DNAse I and buffer needed, and brought up to a final volume
of 10 μl per sample with sterile H2O. After adding the DNAse
mix, the samples were heated at 37°C for 30 min. Then, for
each sample, 1 μl of 50 mM EDTA was added and incubated at
65°C for 10 min to inactivate it. First-strand cDNA was
synthesized using the SuperScriptTM IV First-Strand Synthesis
System (Invitrogen) kit and 2 μl of RNA according to the
manufacturer’s instructions. Primer3Plus software1 was used
to design PCR primers with G/C content and the presence of
introns (Supplementary Table 1). A volume of 1 μl of cDNA
products were amplified with 12.5 μl of Green Taq polymerase,
1 μl of each specific primer, and 9.5 μl of H20 milliQ. A
5-min cycle at 94°C was followed by 60 s at 94°C, 30 s at 60°C,
and 1 min at 72°C for 35 cycles, and then another 5-min cycle
at 72°C. A no-sample negative control was always included in
each set of reactions. PCR products were loaded onto 1%
1
Extraction and quantification of
centellosides
Centellocide production was determined according to
Hidalgo et al. (2016) and Alcalde et al. (2022) with slight
modification. We weighted 0.5 g of freeze dry material (DW)
of hairy roots and added 10 ml of methanol: H2O (9:1)
suspension, which was sonicated for 1 h at room temperature.
The following step was to centrifuge at 20,000 rpm for
10 min. After separating the supernatant, the previous step
was repeated. The supernatants of the various samples were
placed on porcelain mortars and evaporated at 38°C for
approximately 24 h before being redissolved in 1.5 ml of
methanol. The methanolic extract was filtered through a
0.22 μm filter for HPLC quantification of centellosides. The
HPLC system consisted of a Waters 600 Controller pump, a
Waters 717 Autosampler automatic injector, a Jasco variable
length (UV) 1570 detector, and Borwin data analysis software
version 1.5. At room temperature, a Lichrospher 100 RP18
5 μm column (250 × 0.4 mm) was used for gradient
chromatography, as described in Supplementary Table 2. The
mobile phase consisted of acetonitrile and ammonium
phosphate (10 mM), which had been acidified with orthophosphoric acid to a pH of 2.5. The acidification improved
the definition of the compound peaks. The flow rate was
1 ml/min, and the injection volume was 10 μl. The detector
wavelength was set to 214 nm, 1.00 au/v, and the run
time was 45 min. To quantify the centellosides (asiatic acid,
madecassic acid, asiaticoside, and madecasoside), standards
of these 4 compounds were used to prepare calibration
curves at concentrations of 10, 25, 50, 100, 250, and
500 ppm.
https://www.bioinformatics.nl/cgi-bin/primer3plus/primer3plus.cgi
(Accessed August 22, 2022).
Frontiers in Plant Science
03
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
United States.2 Data are presented as the mean ± standard
deviation. For statistical comparison, a multifactorial ANOVA
analysis was performed, followed by Tukey’s multiple comparison
test. For morphologic traits, phytohormone concentration, and
centelloside production, a p-value ≤ 0.05 was assumed to indicate
a significant difference.
Machine learning models, principal component analysis
(PCA), and Pearson correlation were performed using R Statistical
Software (R Core Team, 2021). The caret package (Kuhn, 2021)
was used to create LDA, SVM, RF, and ANN models, whereas
randomForestSRC package (Ishwaran and Kogalur, 2022) was
used to execute multivariate multiple regression models. A general
scheme of the modeling process is presented in
Supplementary Figure 9. For model validation, data split and
repeated cross-validation methods were used, with 10-fold crossvalidation repeated 5 times. Accuracy or coefficient of
determination (R-squared) was used as metrics to evaluate the
performance of each model, along with specificity and sensitivity.
Factoextra (Kassambara and Mundt, 2020) package was used for
the PCA, while corrplot package (Wei and Simko, 2021) was used
for the calculation of the Pearson correlation coefficient (r). The
multivariate multiple regression model’s decreased accuracy
values were used to perform a hierarchical clustering analysis and
were displayed as a heatmap. Datasets for model development and
PCA analysis were preprocess by autoscaling method.
FIGURE 1
Average morphology of the different transformed hairy roots of
C. asiatica at different stages of development.
Hormonal profiling of hairy root lines
Samples were collected for hormonal profiling and
immediately frozen in liquid nitrogen before being stored at
−80°C for subsequent analyses. The endogenous
concentrations of the compounds: abscisic acid (ABA),
salicylic acid (SA), jasmonates (12-oxo-phytodienoic acid
[OPDA], jasmonic acid [JA], and jasmonoyl isoleucine
[JA-Ile]), cytokinins (2-isopentenenyladenine [2iP], IPA,
trans-zeatin [t-Z], and trans-zeatin riboside [t-ZR]), the
auxin indole 3-acetic acid (IAA), and gibberellins (GA1, GA4,
and GA7) were quantified using a protocol modified by
Müller and Munné-Bosch (2011). For each hairy root line,
100 ± 5 mg samples were placed in liquid nitrogen in a 2 ml
Eppendorf tube using the mixer mill MM400 (Retsch GmbH,
Haan, Germany), and then extracted twice with extraction
solvent (methanol:isopropanol:acetic acid in a proportion
50:49:1 [v/v/v] with 1% of glacial acetic acid) using
ultrasonication (4–7°C). Deuterium-labeled compounds
(Olchemim, Olomuc, Czech Republic) were used as internal
standards for all phytohormones to estimate recovery rates
for each sample. The quantifications were performed by
preparing a calibration curve with each of the 13 compounds
analyzed and calculating the compound/standard ratio using
Analyst™ software (Applied Biosystems, Inc., Foster City,
CA, United States). The results were expressed using the FW
of the samples.
Feeding experiment
To validate the model predictions and multivariate analysis
results from the previous sections, we designed an experiment that
consisted of supplying for 14 and 28 days ABA (13 and 1,300 ng/l)
to the line L1 and IPA (1.5 and 150 ng/l) to the line L3. We used a
single root to maximize visualization of the effect on growth and
branching rates. ABA was bought to Sigma-Aldrich (Steinheim
am Albuch, Germany) and IPA was bought to Cayman chemical
(Ann Arbor, Michigan, USA). The stock solutions where prepared
at 1 mg/ml in methanol followed by serial dilution prior to
be added to culture media.
Results
Hairy root traits and centelloside content
Statistical analysis and machine learning
models
The plant material used in this study consisted of 10 hairy root
lines free of Agrobacterium rhizogenes (L1, L2, L3, L4, L6, L7, L8,
L10, L12, and L14), which were cultivated for 2 and 4 weeks,
before morphological traits and biomass production were
registered, as described in the Material and methods section.
Figure 1 shows the development of the root lines throughout the
The statistical analysis was carried out using GraphPad Prism
version 6.04 for Windows, GraphPad Software, La Jolla California,
2
Frontiers in Plant Science
04
www.graphpad.com (Accessed August 22, 2022).
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
experiment. Most of the lines had the typical morphology of hairy
roots, but there were noticeable differences in branching rate,
growth rate, and biomass accumulation. Statistical differences
were found between lines in terms of branching and growth rate,
but there were no significant changes between sampling times.
Figure 2A shows the branching rate, with L1 standing out from
the others, followed by L12. L3 had the lowest branching values at
2 and 4 weeks of growth. In terms of growth rate (Figure 2B), two
lines predominated: L1 and L2. Line L14 has the lowest growth
rate value at the end of 4 weeks. Finally, we observed significant
differences in biomass productivity (Figure 2C) between weeks 2
and 4 for lines L1, L2, L8, L10, and L12, with increases ranging
from 2 to 5 times.
The total productivity of centelloside was expressed in two
ways: mg per g of dry weight (DW) and mg per liter of culture
medium (Figure 3). These results showed a similar profile
between hairy root lines as the traits mentioned previously, as
well as significant differences between sampling times. Lines 1
and 2 had the highest production values, followed by lines 10,
12, and 14. Lines 3 to 8 had the lowest values. The productivity
range of centelloside values in mg/g DW oscillated between
0.14 ± 0.02 and 0.96 ± 0.03 after 14 days, and between 0.44 ± 0.27
and 5.49 ± 0.20 after 28 days. These results revealed the root
lines’ various capacities to accumulate this type of compound
over time. Some of them, such as L2, showed a content that was
8 times higher at 28 days than it was at 14 days, whereas line L7’s
increase was only 0.7 times. In terms of centellosides profile,
made cassoside was the main compound in almost all lines
(Supplementary Figure 2). We can see from these results that
the transformed root lines can be divided into at least 3 groups:
L1 and L2, then L3, L4, L6, L7, and L8, and finally L10, L12,
and L14.
Gene expression
PCR (data not shown) and semi-quantitative RT-PCR were
used to confirm the integration and expression of the rol and aux
genes. Supplementary Figure 3 shows the results of semiquantitative RT-PCR in 10 C. asiatica hairy root lines grown in MS
basal medium after 28 days, with the 5.8 s rRNA used as a
housekeeping gene for normalization. A principal component
analysis (PCA) was used to investigate how the expression of these
genes was related to hairy root traits and centelloside content.
Figure 4A summarizes the information about hairy root samples
and their multiple gene expression by two components:
PC1 = 63.5% and PC2 = 27.5%, which account for 91% of the
model’s total variance. L3, L4, L6, and L7 are the lines with the
lowest expression of all genes, according to PC1. According to
PC2, other subgroups can be seen within the lines with higher
expression. The first (L10, L12, and L14) was associated with high
expressions of the rolC and rolB genes, while the second (L1, L2,
and L8) showed higher expressions of the rolA and aux1 genes. A
positive correlation was observed between centelloside content,
branching, biomass productivity, rol, and aux1 genes (Figure 4B).
The highest centelloside productions were strongly related to rolA
(r = 0.71) and aux1 (r = 0.70) genes. The rolC was the least
effective, with a slightly negative effect on elongation rates
(r = −0.29). Similar behavior was observed for samples at 14 days
(Supplementary Figure 4).
A
B
C
Prediction of production degree based
on gene expression
The statistically significant correlations and differences
established the concept of identifying production lines based on
their gene expression profile. We assigned the following tags to the
subgroups that represent the centelloside content: HIGH (L1 and
L2), MID (L10, L12, and L14), and LOW (L3, L4, L6, L7, and L8).
Four different classification machine learning algorithms were
tested on a dataset (Supplementary Table 3) containing gene
expression information from 10 lines cultivated for 14 and 28 days,
and accuracy was used to track the models’ performance (Table 1).
FIGURE 2
Morphological traits of hairy root lines of C. asiatica.
(A) Branching rate. (B) Growth rate. (C) Biomass productivity.
Data represent the mean±SD of three replicates. Different letters
show significant differences between hairy root lines. ns=no
differences (α =0.05).
Frontiers in Plant Science
05
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
Hormone profiles
The most accurate model was random forest (RF) with 91.33%
correct classification, and the least accurate was artificial neural
network (ANN) with 89.5% correct classification. The sensitivity
range was 62.5 to 100%, with RF having the highest values,
followed by linear discriminant analysis (LDA) and ANN, and
supported vector machine (SVM) having the lowest values. Model
specificity ranged between 78.6 and 100%. The importance of each
gene for correct classification was calculated and ranked by the
mean decrease in Gini. The rolC is at the top of the list, followed
by rolA, rolB, and finally aux1 (Supplementary Figure 5A).
The hormone profile of the hairy root lines was evaluated at
14 and 28 days, with 13 compounds measured, including ABA,
SA, OPDA, JA, JA-Ile, 2iP, IPA, t-Z, t-ZR, IAA, GA1, GA4, and
GA7. GA4 was detected, but GA1 and GA7 values were below the
detection threshold for all rhizoclones (Supplementary Figure 6).
LOW centelloside producer lines had higher concentrations of
ABA, SA, JA-Ile, and IAA, especially at 4 weeks of growth. MID
centelloside producer lines had the highest concentration of 2-iP
throughout the experiment. HIGH producer centelloside lines
had the highest concentration value for IPA. The other hormones
(JA, OPDA, t-Z, t-ZR, and GA4) showed different values
depending on the week of growth for each transformed root line.
The loadings plot in Figure 5 depicts the relationship between
hairy root traits, centelloside contents, and hormone profile at
various sampling times. The PCA was composed by nine
components, where PC1 covered the 56% and PC2 = 22.7% which
account for 78.9% of the model’s total variance. At 14 days
(Figure 5A; Supplementary Figure 7), the only hormone with a
strong positive correlation (r > 0.7) for all traits and centelloside
content was IPA hormone. OPDA showed a positive effect for
branching and centelloside content, and t-Z only showed a
positive correlation for centelloside content. IAA, on the other
hand, had a strong negative effect (r < 0.7) on centelloside
production, branching, and biomass production. ABA also had a
negative effect on centelloside production and branching. IPA was
strongly correlated (r > 0.8) with hairy root elongation and had a
positive effect on centelloside content and biomass productivity at
FIGURE 3
Centelloside content in mg/g DW of hairy root lines of C. asiatica
measured as the sum of asiaticoside + madecassoside + asiatic
acid + madecassic acid. Data represent the mean±SD of three
replicates. Different letters show significant differences between
hairy root lines. ns=no differences (α =0.05).
A
B
FIGURE 4
Gene expression analysis and correlation with morphological traits and centelloside production of hairy roots at 28days of culture. (A) Biplot of
Principal Component Analysis of the genes studied. (B) Pearson’s correlation analysis of gene expression, morphological traits, and centelloside
production. PWD=centelloside production in mg/g, PL=centelloside production in mg/L, Elongation=growth rate, and BioMP=biomass
productivity.
Frontiers in Plant Science
06
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
TABLE 1 Prediction of production degree based on gene expression
and hormone profiles.
Model Accuracy
%
Sensitivity %
purpose using the randomForestSRC R-package (see Materials
and methods). The different hormones were treated as
dependent variables in the model, while the genetic expression
of rolA, rolB, rolC, and aux1 genes was treated as independent
variables. The R-squared value for the model with all
hormones was 0.788, which was used to measure the goodness
of fit, the performance error of the model was 0.1998. The
variable importance (VIMP) was used to compare the
influence of genes on each hormone profile to aid in the
interpretation of the multivariate regression model. Figure 6
shows a cluster analysis of hormones based on the influence
of gene expression. ABA was most influenced by the behavior
of rolB, followed by rolA and rolC, and aux1 had a low
influence on its content. IPA was heavily influenced by aux1,
followed by rolA. All genes had a greater than average
influence on the 2 iP, with rolC having the greatest effect. The
GA4 behavior was slightly affected by rolA and aux1. The
genes rolA and rolC had a positive impact on IAA, while aux1
and rolB had an average impact. Starting with SA, gene
influence declined, followed by JA-ile, t-Z, OPDA, and JA as
the hormone least influenced by gene expression. To improve
the model’s fitness, the less influenced hormones by the rol or
aux genes were eliminated one by one. When JA, OPDA, t-Z,
JA-ile, SA, and GA4 were excluded, the best fitness was 0.837.
In contrast, the absence of ABA resulted in the greatest
decrease in model fitness, with an R-squared value of less than
0.3, followed by IPA and 2 iP. This abrupt decrease in
R-squared value confirms the connection between the
expression of these genes and the hormone profile of ABA,
IPA, and 2 iP.
Y-randomization was implemented to prove accurate
prediction potential of the model. We selected the top two most
influenced hormones ABA, IPA and JA which was the lowest
influenced by the gene expression to do this test. Individual
models were built for each of these three hormones, the R-squared
value was calculated and compared against the population of
R-squared values obtained after 1,000 permutation
(Supplementary Figure 10). As a result of this test, ABA and IPA
showed to be accurate predicted by the regression model.
Additionally, JA showed the overlapping of the simulated values
with the original value of R-squared. The above results matched
with the multivariate multiple regression model
previously developed.
Specificity %
HIGH MID LOW HIGH MID LOW
Prediction of production degree based on gene expression
LDA
90.77
66.67
100
100
100
100
83.33
SVM
90.05
83.33
100
62.5
78.57
100
91.67
RF
91.33
83.33
100
87.5
92.9
100
91.7
ANN
89.5
66.67
100
100
100
100
83.33
Prediction of production degree based on hormone profiles
LDA
82.57
100
100
50
75
100
100
SVM
100
100
100
100
100
100
100
100
100
100
100
100
100
100
93.75
100
100
100
100
100
100
RF
ANN
28 days (Figure 5B; Supplementary Figure 8). OPDA and GA4
showed a negative correlation (r < 0.73) for all traits and
centelloside content. ABA, SA, and t-Z were also negatively
correlated except for root elongation. When the hormone content
is compared between the two sampling times, the high amount of
IPA hormone is repeatedly associated with a high content of
centelloside, while the high amount of ABA is associated with
low content.
Prediction of production degree based
on hormone profiles
Similarly, we investigated whether the hormonal profiles of
transformed root lines could predict the degree of centelloside
production. Four different classification machine learning
algorithms were tested on a dataset (Supplementary Table 4)
containing hormone content from the HIGH, MID, and LOW
groups, and accuracy was used to track the models’ performance
(Table 1). The most accurate models were RF and SVM, with 100%
correct classification, and the least accurate model was LDA, with
82.57%. The sensitivity ranged from 50 to 100%, with SVM, RF,
and ANN having the highest values and LDA having the lowest.
The specificity range of the models was between 75 and 100%. The
SVM model performed the best in classifying different degrees of
centelloside production. The importance of each hormone for
correct classification was calculated and ranked by the mean
decrease in the Gini coefficient. The top 5 on this list were IPA, 2iP,
ABA, IAA, and GA4 (Supplementary Figure 5B).
Feeding experiments
Figure 7A shows the development of root lines under the
exogenous hormonal influence of ABA or IPA. The controls
behaved consistently with the previous experiments, whereas the
effects of ABA were perceived after 14 days at concentrations of
13 ng/l, causing significant decreases in the measured traits
(Figures 7B–D) on the HIGH line. After 28 days, the lower
concentration reduced branching, growth rate, and biomass
Gene influence on hormone profiles
We investigated how genes influence hormone profiles
after studying how traits and centelloside content can
be predicted by their gene expression or hormone profile. A
multivariate multiple regression model was created for this
Frontiers in Plant Science
07
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
A
B
FIGURE 5
Loading plot of hormonal profiling, morphological traits and centelloside production of hairy roots cultures (A) at 14days and (B) at 28days.
PWD=centelloside production in mg/g, PL=centelloside production in mg/L, Elongation=growth rate, and BioMP=biomass productivity.
FIGURE 6
Global heatmap of the influence of rol and aux1 genes on the hormone profile in hairy roots of C. asiatica.
productivity by 0.83, 0.54, and 0.45 times, respectively compared
to the control. At 1,300 ng/l, the same traits were reduced 0.65,
0.24, and 0.28 times, respectively. When IPA was given to the
LOW line, the opposite effect was seen; the effects on traits were
visible at 28 days and concentrations of 150 ng/l. At the highest
concentration, branching, growth rate, and biomass productivity
were increased 2.60, 1.25, and 3.8 times after 28 days, respectively.
Branching rate and biomass productivity were the most affected
by ABA and IPA. Regarding centelloside content
(Supplementary Figure 11), after 28 days on ABA treatment the
HIGH line showed a decrease in 21% compared to the control
when ABA was at 13 ng/l, while 74% of control when ABA was at
Frontiers in Plant Science
1300 ng/l. In contrast, the LOW line after 28 days on IPA treatment
showed an increase in 1.76 times compared to the control when
IPA was at 1.5 ng/l, while 3.21 times when IPA was 150 ng/l.
Discussion
The hairy root syndrome is a disease that affects many plants
and is caused by the infection and subsequent insertion of a
fragment of the A. rhizogenes plasmid known as T-DNA
(“transfer” DNA; Mauro et al., 2017). The rol genes, which are
found in the TR-DNA region, are primarily responsible for the
08
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
morphology and formation of these hairy roots (Ozyigit
et al., 2013).
This type of in vitro culture system is useful for secondary
metabolites biosynthesis and biotechnological production (Bonfill
et al., 2015) since these cultures grow much faster than other types
of in vitro cultures (Kim et al., 2007) and produce a spectrum of
secondary metabolites similar to plant roots (Ruslan et al., 2012).
We observed the presence and expression of all rol and aux
genes in all hairy lines studied (Supplementary Figure 3), which is
similar to the work of Komarovská et al. (2009) in Hypericum
tomentosum. and Hypericum tetrapterum, but differs from what
was observed by Alpizar et al. (2008) in Coffea arabica, where the
presence of any aux gene was found in all transformed lines.
We selected the A. rhizogenes TL-DNA genes rolA, B and C
because they have been shown to play the most relevant role in
hairy root development (Sarkar et al., 2018; Bahramnejad et al.,
2019), and the aux1 gene as a representative of the proper
integration of the TR-DNA region in the transformed root lines
(Mano and Nemoto, 2012). In our study, the expression level of all
rol genes was higher than that of the aux gene
(Supplementary Figure 3), which may be since only the presence
of TL-DNA genes is required for long-term hairy root growth
(Dessaux and Petit, 1994; Chriqui et al., 1996). The rolD was not
analyzed because it was not detected in all A. rhizogenes strains
(Pavlova et al., 2014).
When we compared all the hairy root lines or rhizoclones,
we observed differences in growth and morphology that could
be grouped into different categories, and it was thanks to the
cultivation of these roots in a solid medium that the quantification
of their traits was easily reproducible and simple to do. These
differences between rhizoclones were attributed to variations in
the nature, size, and number of T-DNA integrations into the host
genome (Alpizar et al., 2008; Roychowdhury et al., 2015; Thwe
et al., 2016). An evaluation with new methods to determine the
number of copies of T-DNA integrated in the root lines could
be an extension of the present study (Głowacka et al., 2016), which
would allow identifying if the number of integrated copies is more
relevant than the place of integration of the genes rol or aux1.
HIGH group lines had the highest expression value of the rolA
gene, as well as the highest rooting rate, growth rate, biomass
production, and centelloside content (Supplementary Figure 3).
This gene is found in all Ri-plasmids and may be involved in the
interaction with nucleic acids, which may be related to the
regulation of gene expression in plants (Pavlova et al., 2014). The
A
B
C
D
FIGURE 7
Feeding experiments on hairy roots lines, where L1 represent the HIGH group, while L3 the LOW group. (A) Developmental stages at different
hormone concentration. (B) Branching rate at 14 and 28days. (C) Growth rate. (D) Productivity of biomass. Data represent the mean±SD of three
replicates.
Frontiers in Plant Science
09
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
importance of this gene in secondary metabolism production was
observed in Rubia cordifolia callus for anthraquinone production
(Shkryl et al., 2008).
The evaluation of the gene expression using a multivariable
analysis such as PCA allowed us to visualize in a simplified way
the correlation and behavior of each of the characteristics of the
different root lines (centellosides content, branching, elongation,
and biomass production) over time. This analysis identified the
grouping of root lines based on the degree of specific and recurrent
gene expression. In general, low expression of rol and aux1 genes
was associated with poor traits and low centelloside production.
The genes aux1 and rolA were found to be more closely linked to
lines with higher centelloside and trait content values. The aux1
gene is responsible for differences in hairy root growth and
morphology (Ozyigit et al., 2013). This could be related to the
line’s high rooting and growth rate in long-term culture
(Figures 2, 4A), which was also observed by Lütken et al., 2017.
MID group lines had the highest levels of rolB and rolC gene
expression (Figure 4A). The rolB is the most powerful inducer of
secondary metabolism and the greatest suppressor of cell growth
(Bulgakov, 2008). The MID group lines produced lower amount
of centellosides than the HIGH group lines, which could
be attributed to the high expression of rolC since it has previously
been shown to have antagonistic effects between these two genes
(Bulgakov et al., 2003). The rolC gene can stimulate the production
of tropane alkaloids (Bonhomme et al., 2000), pyridine alkaloids
(Palazon et al., 1998), ginsenosides (Bulgakov et al., 1998), and
flavonoids (Ismail et al., 2017) in different in vitro culture systems,
which differs somewhat from our studies.
The multivariable analysis (PCA) also exposed the behavior
of the different hormones between traits and centelloside
production, revealing a positive correlation with the IPA
hormone content and a negative correlation with the ABA
hormone. It is well known that ABA regulates numerous aspects
of plant growth. Dicot plants deficient in this hormone have
reduced seed dormancy and wilty phenotypes (Harris, 2015;
Nambara, 2016). However, high levels of ABA have been shown
to inhibit cell division in apical meristems and root elongation
(Bai et al., 2009; Takatsuka and Umeda, 2014; Yang et al., 2014;
Sun et al., 2018). This last scenario, in which high content is
negatively correlated with elongation and branching, is consistent
with our results, as is the low production of triterpene saponins.
In terms of IPA, its high presence in hairy root lines was
consistent with improved biomass production, elongation, and
branching, and is supported by the abundance of lateral root
meristems, which are one of the main sites of cytokinin synthesis
(Nordström et al., 2004).
We focus on centelloside production for machine learning
modeling since it is one of the most important parameters for
biotechnological applications. The correlation analysis allowed us
to consider the positive relationship between centelloside
production and biomass production, elongation, and branching
rate, allowing us to omit them from the models and simplify their
execution. By discretizing the content of centellosides, it was
Frontiers in Plant Science
possible to apply different supervised machine learning models,
such as dimensionality reduction (LDA; Zhao et al., 2020),
instance-based (SVM; Cristianini and Ricci, 2008), ensemble
methods (RF; Liaw and Wiener, 2002), and artificial neural
network (ANN; Venables and Ripley, 2002). The reason for
selecting these models was due to their nature, as mentioned
above. In general, RF and SVM performed the best with the data
presented in this work, correctly classifying the samples into the
classes proposed. The LDA, on the other hand, produced the worst
results, which could be attributed to the fact that the data
distribution was not normal for all variables. The data distribution
was identified by Shapiro–Wilk test and the results are shown in
Supplementary Table 5.
Multiple multivariate regression models were used to
understand how certain hormones and gene expression levels (rol
and aux1 genes) interact to coordinate root growth and
development. This allowed us to simultaneously evaluate the
influence of each level of genetic expression on the profile of each
hormone studied. The random forest method has the advantage
of being able to work with data whose distribution may or may
not be normal; it evaluated the importance of each variable
within the model, allowing identifying the degree of influence of
genes on each of the growth regulators. Feeding experiments
validated the model by demonstrating that the analyzed
phytohormones (IPA and ABA) were determinants in increasing
the high producer line and decreasing the low producer line, as
well as influencing centelloside production.
The use of this biotechnological platform together with
machine learning techniques resulted in the implementation of
models that allow us to discriminate root lines based on their
level of production of secondary metabolites such as
centelloside, with random forest outperforming all others. This
discrimination was made possible by using gene expression
levels of the rol and aux1 genes, as well as hormone profiles.
Furthermore, the degree of influence of each gene on the
individual profile of each hormone studied was determined,
with IPA and ABA being the most affected due to the action of
the rol and aux1 genes. Finally, the results of the gene influence
analysis on these two specific hormones were successfully tested
with feeding experiments aimed at reversing the effect on root
morphology and centelloside content.
Data availability statement
The original contributions presented in the study are included
in the article/Supplementary material, further inquiries can
be directed to the corresponding authors.
Author contributions
MB, JP, and DH-M designed the research. MA, MM, and
SM-B determined the hormone profile. MA, ML, and PG
10
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
build neural networks models. MA and DH-M build the
others machine learning models. MA performed tissue
culture and metabolites determination. All authors
contributed to the article and approved the submitted
version.
that could be construed as a potential conflict of
interest.
Publisher’s note
All claims expressed in this article are solely those of the
authors and do not necessarily represent those of their affiliated
organizations, or those of the publisher, the editors and the
reviewers. Any product that may be evaluated in this article, or
claim that may be made by its manufacturer, is not guaranteed or
endorsed by the publisher.
Funding
This work was partially funded by the Spanish Ministry of
Science and Innovation, with project number PID2020113438RB-I00, and by the Catalan Government, with project
number 2017SGR980.
Supplementary material
Conflict of interest
The Supplementary material for this article can be found online
at: https://www.frontiersin.org/articles/10.3389/fpls.2022.1001023/
full#supplementary-material
The authors declare that the research was conducted
in the absence of any commercial or financial relationships
References
Alcalde, M. A., Cusido, R. M., Moyano, E., Palazon, J., and Bonfill, M. (2022).
Metabolic gene expression and centelloside production in elicited Centella asiatica
hairy root cultures. Ind. Crop. Prod. 184:114988. doi: 10.1016/J.INDCROP.
2022.114988
Cardillo, A. B., Giulietti, A. M., Palazón, J., and Bonfill, M. (2013). Influence of hairy
root ecotypes on production of tropane alkaloids in Brugmansia candida. Plant Cell
Tissue Organ Cult. 114, 305–312. doi: 10.1007/S11240-013-0326-Y/FIGURES/5
Chen, S. L., Yu, H., Luo, H. M., Wu, Q., Li, C. F., and Steinmetz, A. (2016).
Conservation and sustainable use of medicinal plants: problems, progress, and
prospects. Chinese Med. (United Kingdom) 11, 1–10. doi: 10.1186/S13020-016-01087/TABLES/5
Alpizar, E., Dechamp, E., Lapeyre-Montes, F., Guilhaumon, C., Bertrand, B.,
Jourdan, C., et al. (2008). Agrobacterium rhizogenes-transformed roots of coffee
(Coffea arabica): conditions for Long-term proliferation, and morphological and
molecular characterization. Ann. Bot. 101, 929–940. doi: 10.1093/AOB/
MCN027
Chriqui, D., Guivarc’h, A., Dewitte, W., Prinsen, E., and Van Onkelen, H. (1996).
Rol genes and root initiation and development. Plant Soil 187, 47–55. doi: 10.1007/
BF00011656
Arribas-López, E., Zand, N., Ojo, O., Snowden, M. J., and Kochhar, T. (2022). A
systematic review of the effect of Centella asiatica on wound healing. Int. J. Environ.
Res. Public Health 19, 3266. doi: 10.3390/ijerph19063266
Cristianini, N., and Ricci, E. (2008). “Support vector machines,” in Encyclopedia
of Algorithms. ed. M. Y. Kao (Boston, MA: Springer US), 928–932. doi:
10.1007/978-0-387-30162-4_415
Bahramnejad, B., Naji, M., Bose, R., and Jha, S. (2019). A critical review on use of
agrobacterium rhizogenes and their associated binary vectors for plant
transformation. Biotechnol. Adv. 37:107405. doi: 10.1016/J.BIOTECHADV.
2019.06.004
Dessaux, Y., and Petit, A. (1994). “Opines as screenable markers for plant
transformation,” in Plant Molecular Biology Manual. eds. B. Gelvin and A. S.
Robbert (Netherlands: Springer), 171–182.
Bai, L., Zhang, G., Zhou, Y., Zhang, Z., Wang, W., Du, Y., et al. (2009). Plasma
membrane-associated proline-rich extensin-like receptor kinase 4, a novel
regulator of Ca2+ signalling, is required for abscisic acid responses in
Arabidopsis thaliana. Plant J. 60, 314–327. doi: 10.1111/J.1365-313X.2009.
03956.X
Dhillon, P., Kumari, P., Kumari, C., and Singh, P. S. (2017). Phytochemical
screening of selected medicinal plants for secondary metabolites. J. Life. Sci. Sci. Res
3, 1151–1157. doi: 10.21276/ijlssr.2017.3.4.9
Dilshad, E., Noor, H., Nosheen, N., Gilani, S. R., Ali, U., and Khan, M. A. (2021).
“Influence of rol Genes for Enhanced Biosynthesis of Potent Natural Products,” in
Chemistry of Biologically Potent Natural Products and Synthetic Compounds. (USA:
John Wiley & Sons, Ltd), 379–404.
Bonfill, M., Cusidó, R. M., Lalaleo, L., and Palazón, J. (2015). “Plant cell and
organ cultures as a source of phytochemicals,” in Recent Advances in
Pharmaceutical Sciences. eds. D. Muñoz-Torrero, M. P. Vinardell and J. Palazón
(Kerala, India: Research Signpost), 33–49.
Donini, M., and Marusic, C. (2018). “Hairy roots as bioreactors for the production
of biopharmaceuticals,” in Hairy Roots: An Effective Tool of Plant Biotechnology. eds.
S. Vikas, M. Shakti and M. Sonal (Singapore: Springer), 213–225.
Bonhomme, V., Laurain-Mattar, D., and Fliniaux, M. A. (2000). Effects of the rol
C gene on hairy root: induction development and Tropane alkaloid production by
Atropa belladonna. J. Nat. Prod. 63, 1249–1252. doi: 10.1021/NP990614L
Gallego, A., Ramirez-Estrada, K., Vidal-Limon, H. R., Hidalgo, D., Lalaleo, L., Khan
Kayani, W., et al. (2014). Biotechnological production of centellosides in cell cultures
of Centella asiatica (L) Urban. Eng. Life Sci. 14, 633–642. doi: 10.1002/ELSC.201300164
Bulgakov, V. P. (2008). Functions of rol genes in plant secondary metabolism.
Biotechnol. Adv. 26, 318–324. doi: 10.1016/J.BIOTECHADV.2008.03.001
Głowacka, K., Kromdijk, J., Leonelli, L., Niyogi, K. K., Clemente, T. E., and
Long, S. P. (2016). An evaluation of new and established methods to determine
T-DNA copy number and homozygosity in transgenic plants. Plant Cell Environ. 39,
908–917. doi: 10.1111/pce.12693
Bulgakov, V. P., Khodakovskaya, M. V., Labetskaya, N. V., Chernoded, G. K., and
Zhuravlev, Y. N. (1998). The impact of plant rolC oncogene on ginsenoside
production by ginseng hairy root cultures. Phytochemistry 49, 1929–1934. doi:
10.1016/S0031-9422(98)00351-3
Gutierrez-Valdes, N., Häkkinen, S. T., Lemasson, C., Guillet, M.,
Oksman-Caldentey, K. M., Ritala, A., et al. (2020). Hairy root cultures—a versatile
tool with multiple applications. Front. Plant Sci. 11:33. doi: 10.3389/
FPLS.2020.00033/BIBTEX
Bulgakov, V. P., Tchernoded, G. K., Mischenko, N. P., Khodakovskaya, M. V.,
Glazunov, V. P., Radchenko, S. V., et al. (2002). Effect of salicylic acid, methyl
jasmonate, ethephon and cantharidin on anthraquinone production by Rubia
cordifolia callus cultures transformed with the rolB and rolC genes. J. Biotechnol. 97,
213–221. doi: 10.1016/S0168-1656(02)00067-6
Harris, J. M. (2015). Abscisic acid: hidden architect of root system structure.
Plants 4, 548–572. doi: 10.3390/PLANTS4030548
Bulgakov, V. P., Tchernoded, G. K., Mischenko, N. P., Shkryl, Y. N., Glazunov, V. P.,
Fedoreyev, S. A., et al. (2003). Effects of Ca2+ channel blockers and protein kinase/
phosphatase inhibitors on growth and anthraquinone production in Rubia cordifolia
callus cultures transformed by the rolB and rolC genes. Planta 217, 349–355. doi:
10.1007/S00425-003-0996-5/TABLES/2
Frontiers in Plant Science
Hashemi, S. M., and Naghavi, M. R. (2016). Production and gene expression of
morphinan alkaloids in hairy root culture of Papaver orientale L. using abiotic
elicitors. Plant Cell Tissue Organ Cult. 125, 31–41. doi: 10.1007/S11240-015-0927-8/
FIGURES/8
11
frontiersin.org
Alcalde et al.
10.3389/fpls.2022.1001023
Hidalgo, D., Steinmetz, V., Brossat, M., Tournier-Couturier, L., Cusido, R. M.,
Corchete, P., et al. (2016). An optimized biotechnological system for the production
of centellosides based on elicitation and bioconversion of Centella asiatica cell
cultures. Eng. Life Sci. 17, 413–419. doi: 10.1002/ELSC.201600167
Improvement: New Approaches and Modern Techniques. eds. H. K. Rehman, A.
Parvaiz and O. Munir (Boston, MA: Springer US), 1–48.
Palazon, J., Cusido, R. M., Gonzalo, J., Bonfill, M., Morales, C., and Pinol, M. T.
(1998). Relation between the amount of rolC gene product and indole alkaloid
accumulation in Catharanthus roseus transformed root cultures. J. Plant Physiol.
153, 712–718. doi: 10.1016/S0176-1617(98)80225-3
Ishwaran, H., and Kogalur, U. B. (2022). Fast unified random forests for survival,
regression, and classification (RF-SRC). Available at: https://cran.r-project.org/
package=randomForestSRC (Accessed August 22, 2022).
Palazón, J.., Cusidó, R. M.., Roig, C.., and Pinol, M. T. (1997). Effect of rol genes
from agrobacterium rhizogenes TL-DNA on nicotine production in tobacco root
cultures. Plant Physiol. Biochem. 35, 155–162. Available at: https://agris.fao.org/
agris-search/search.do?recordID=FR9702596 (Accessed April 18, 2022).
Ismail, H., Dilshad, E., Waheed, M. T., and Mirza, B. (2017). Transformation of
lettuce with rol ABC genes: extracts show enhanced antioxidant, analgesic, antiinflammatory, antidepressant, and anticoagulant activities in rats. Appl. Biochem.
Biotechnol. 181, 1179–1198. doi: 10.1007/S12010-016-2277-3/FIGURES/7
Pavlova, O. A., Matveyeva, T. V., and Lutova, L. A. (2014). rol-genes of
agrobacterium rhizogenes. Russ. J. Genet. Appl. Res 4, 137–145. doi: 10.1134/
S2079059714020063
Joshi, K., and Chaturvedi, P. (2013). Therapeutic efficiency of Centella asiatica (L.)
Urb. An underutilized green leafy vegetable: an overview. Int. J. Pharma Bio Sci. 4,
135–149. Available at: https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.
1.448.1137&rep=rep1&type=pdf (Accessed August 12, 2022).
Prakash, V., Jaiswal, N., and Srivastava, M. (2017). A review on medicinal
properties of Centella asiatica characterization of Endophytic fungi of Mimusops
elengi (Bakul) view project. Artic. Asian J. Pharm. Clin. Res. 10, 69–74. doi:
10.22159/ajpcr.2017.v10i10.20760
Kassambara, A., and Mundt, F. (2020). Factoextra: extract and visualize the results
of multivariate data analyses. Available at: https://cran.r-project.org/
package=factoextra (Accessed August 22, 2022).
Prasad, A., Mathur, A. K., and Mathur, A. (2019). Advances and emerging
research trends for modulation of centelloside biosynthesis in Centella asiatica
(L.) urban: a review. Ind. Crop. Prod. 141:111768. doi: 10.1016/J.INDCROP.
2019.111768
Kim, O. T., Bang, K. H., Shin, Y. S., Lee, M. J., Jung, S. J., Hyun, D. Y., et al. (2007).
Enhanced production of asiaticoside from hairy root cultures of Centella asiatica
(L.) urban elicited by methyl jasmonate. Plant Cell Rep. 26, 1941–1949. doi: 10.1007/
S00299-007-0400-1/FIGURES/6
R Core Team (2021). R: a language and environment for statistical computing.
Available at: https://www.r-project.org/ (Accessed August 22, 2022).
Kiselev, K. V., Dubrovina, A. S., Veselova, M. V., Bulgakov, V. P., Fedoreyev, S. A.,
and Zhuravlev, Y. N. (2007). The rolB gene-induced overproduction of resveratrol
in Vitis amurensis transformed cells. J. Biotechnol. 128, 681–692. doi: 10.1016/J.
JBIOTEC.2006.11.008
Rai, A., Saito, K., and Yamazaki, M. (2017). Integrated omics analysis of
specialized metabolism in medicinal plants. Plant J. 90, 764–787. doi: 10.1111/
TPJ.13485
Komarovská, H., Košnuth, J., Čnellárová, E., and Giovannini, A. (2009).
Agrobacterium rhizogenes-mediated transformation of Hypericum tomentosum L.
and Hypericum tetrapterum fries. Zeitschrift fur Naturforsch. C 64, 864–868. doi:
10.1515/ZNC-2009-11-1218/MACHINEREADABLECITATION/RIS
Roy, A. (2020). Hairy root culture an alternative for bioactive compound
production from medicinal plants. Curr. Pharm. Biotechnol. 22, 136–149. doi: 10.217
4/1389201021666201229110625
Roychowdhury, D., Basu, A., and Jha, S. (2015). Morphological and molecular
variation in Ri-transformed root lines are stable in long term cultures of
Tylophora indica. Plant Growth Regul. 75, 443–453. doi: 10.1007/s10725014-0005-y
Kuhn, M. (2021). Caret: classification and regression training. Available at: https://
cran.r-project.org/package=caret (Accessed August 22, 2022).
Kunjumon, R., Johnson, A. J., and Baby, S. (2022). Centella asiatica: secondary
metabolites, biological activities and biomass sources. Phytomedicine Plus 2:100176.
doi: 10.1016/j.phyplu.2021.100176
Ruslan, K., Selfitri, A. D., Bulan, S. A., Rukayadi, Y., and Elfahmi. (2012). Effect of
agrobacterium rhizogenes and elicitation on the asiaticoside production in cell
cultures of Centella asiatica. Pharmacogn. Mag. 8, 111–115. doi: 10.4103/09731296.96552
Liaw, A., and Wiener, M. (2002). Classification and regression by randomForest.
R News 2, 18–22. Available at: https://cran.r-project.org/doc/Rnews/ (Accessed
August 22, 2022).
Lütken, H., Meropi-Antypa, N., Kemp, O., Hegelund, J. N., and Müller, R. (2017).
“Hairy root cultures of Rhodiola rosea to increase valuable bioactive compounds,”
in Production of Plant Derived Natural Compounds Through Hairy Root Culture ed.
S. Malik (Switzerland: Springer), 65–88.
Sarkar, S., Ghosh, I., Roychowdhury, D., and Jha, S. (2018). “The Effects of rol
Genes of Agrobacterium rhizogenes on Morphogenesis and secondary metabolite
accumulation in medicinal plants”, in Biotechnological Approaches for Medicinal and
Aromatic Plants: Conservation, Genetic Improvement and Utilization, ed. K. Nitish
(Singapore: Springer), 27–51.
Makhzoum, A. B., Sharma, P., Bernards, M. A., and Trémouillaux-Guiller, J.
(2013). Hairy roots: An ideal platform for transgenic plant production and other
promising applications. Phytochem. Plant Growth, Environ. 42, 95–142. doi:
10.1007/978-1-4614-4066-6_6/COVER
Shkryl, Y. N., Veremeichik, G. N., Bulgakov, V. P., Tchernoded, G. K.,
Mischenko, N. P., Fedoreyev, S. A., et al. (2008). Individual and combined effects of
the rolA, B, and C genes on anthraquinone production in Rubia cordifolia
transformed calli. Biotechnol. Bioeng. 100, 118–125. doi: 10.1002/BIT.21727
Mangas, S., Moyano, E., Osuna, L., Cusido, R. M., Bonfill, M., and Palazón, J.
(2008). Triterpenoid saponin content and the expression level of some related genes
in calli of Centella asiatica. Biotechnol. Lett. 30, 1853–1859. doi: 10.1007/
S10529-008-9766-6/FIGURES/5
Sun, L. R., Wang, Y. B., He, S. B., and Hao, F. S. (2018). Mechanisms for abscisic
acid inhibition of primary root growth. Plant Signal. Behav. 13, e1500064–e1500069.
doi: 10.1080/15592324.2018.1500069
Mano, Y., and Nemoto, K. (2012). The pathway of auxin biosynthesis in plants.
J. Exp. Bot. 63, 2853–2872. doi: 10.1093/JXB/ERS091
Sun, B., Wu, L., Wu, Y., Zhang, C., Qin, L., Hayashi, M., et al. (2020). Therapeutic
potential of Centella asiatica and its Triterpenes: a review. Front. Pharmacol.
11:568032. doi: 10.3389/fphar.2020.568032
Matveeva, T., and Otten, L. (2021). Opine biosynthesis in naturally transgenic
plants: genes and products. Phytochemistry 189:112813. doi: 10.1016/J.
PHYTOCHEM.2021.112813
Takatsuka, H., and Umeda, M. (2014). Hormonal control of cell division and
elongation along differentiation trajectories in roots. J. Exp. Bot. 65, 2633–2643. doi:
10.1093/jxb/ert485
Mauro, M. L., Costantino, P., and Bettini, P. P. (2017). The never ending story of
rol genes: a century after. Plant Cell Tissue Organ Cult. 131, 201–212. doi: 10.1007/
S11240-017-1277-5
Thwe, A., Arasu, M. V., Li, X., Park, C. H., Kim, S. J., Al-Dhabi, N. A., et al. (2016).
Effect of different agrobacterium rhizogenes strains on hairy root induction and
phenylpropanoid biosynthesis in tartary buckwheat (Fagopyrum tataricum Gaertn).
Front. Microbiol. 7:318. doi: 10.3389/FMICB.2016.00318/BIBTEX
Müller, M., and Munné-Bosch, S. (2011). Rapid and sensitive hormonal profiling
of complex plant samples by liquid chromatography coupled to electrospray
ionization tandem mass spectrometry. Plant Methods 7, 1–11. doi:
10.1186/1746-4811-7-37/FIGURES/4
Venables, W. N., and Ripley, B. D. (2002). Modern Applied Statistics With S.. New
York: Springer. Available at: http://www.stats.ox.ac.uk/pub/MASS4 (Accessed
August 22, 2022).
Nambara, E. (2016). Abscisic Acid. 2nd ed. Amsterdam, Netherlands: Elsevier
doi:10.1016/B978-0-12-394807-6.00098-8.
Wahby, I., Caba, J. M., and Ligero, F. (2012). Agrobacterium infection of hemp
(Cannabis sativa L.): establishment of hairy root cultures. J. Plant Interact. 8,
312–320. doi: 10.1080/17429145.2012.746399
Nemoto, K., Hara, M., Suzuki, M., Seki, H., Oka, A., Muranaka, T., et al. (2009).
Function of the aux and rol genes of the Ri plasmid in plant cell division in vitro.
Plant Signal. Behav. 4, 1145–1147. doi: 10.4161/PSB.4.12.9904
Wei, T., and Simko, V. (2021). R package “corrplot”: visualization of a correlation
matrix. Available at: https://github.com/taiyun/corrplot (Accessed August 22, 2022).
Nordström, A., Tarkowski, P., Tarkowska, D., Norbaek, R., Åstot, C., Dolezal, K.,
et al. (2004). Auxin regulation of cytokinin biosynthesis in Arabidopsis thaliana: a
factor of potential importance for auxin-cytokinin-regulated development. Proc.
Natl. Acad. Sci. U. S. A. 101, 8039–8044. doi: 10.1073/pnas.0402504101
Yang, L., Zhang, J., He, J., Qin, Y., Hua, D., Duan, Y., et al. (2014). ABA-mediated ROS
in mitochondria regulate root meristem activity by controlling PLETHORA expression
in Arabidopsis. PLoS Genet. 10, e1004791. doi: 10.1371/journal.pgen.1004791
Zhao, H., Lai, Z., Leung, H., and Zhang, X. (2020). “Linear discriminant analysis,”
in Feature Learning and Understanding: Algorithms and Applications (Cham:
Springer International Publishing), 71–85. doi: 10.1007/978-3-030-40794-0_5
Ozyigit, I. I., Dogan, I., and Artam Tarhan, E. (2013). “Agrobacterium rhizogenesMediated Transformation and Its Biotechnological Applications in Crops,” in Crop
Frontiers in Plant Science
12
frontiersin.org