Using machine learning to link the influence of transferred Agrobacterium rhizogenes genes to the hormone profile and morphological traits in Centella asiatica hairy roots

Abstract

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 28 days 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.

Keywords: Agrobacterium rhizogenes; Centella asiatica; centellosides; hairy root cultures; machine learning; plant hormones; random forest.

Figure 1

Average morphology of the different transformed hairy roots of C. asiatica at different stages of development.

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).

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).

Figure 4

Gene expression analysis and correlation with morphological traits and centelloside production of hairy roots at 28 days 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.

Figure 5

Loading plot of hormonal profiling, morphological traits and centelloside production of hairy roots cultures (A) at 14 days and (B) at 28 days. 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.

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 28 days. (C) Growth rate. (D) Productivity of biomass. Data represent the mean ± SD of three replicates.