Phenotyping of Salvia miltiorrhiza Roots Reveals Associations between Root Traits and Bioactive Components

Abstract

Plant phenomics aims to perform high-throughput, rapid, and accurate measurement of plant traits, facilitating the identification of desirable traits and optimal genotypes for crop breeding. Salvia miltiorrhiza (Danshen) roots possess remarkable therapeutic effect on cardiovascular diseases, with huge market demands. Although great advances have been made in metabolic studies of the bioactive metabolites, investigation for S. miltiorrhiza roots on other physiological aspects is poor. Here, we developed a framework that utilizes image feature extraction software for in-depth phenotyping of S. miltiorrhiza roots. By employing multiple software programs, S. miltiorrhiza roots were described from 3 aspects: agronomic traits, anatomy traits, and root system architecture. Through K-means clustering based on the diameter ranges of each root branch, all roots were categorized into 3 groups, with primary root-associated key traits. As a proof of concept, we examined the phenotypic components in a series of randomly collected S. miltiorrhiza roots, demonstrating that the total surface of root was the best parameter for the biomass prediction with high linear regression correlation (R² = 0.8312), which was sufficient for subsequently estimating the production of bioactive metabolites without content determination. This study provides an important approach for further grading of medicinal materials and breeding practices.

Fig. 1.

The workflow for the phenomics study of S. miltiorrhiza roots. (A) Map of trait categories included in this root phenotyping for S. miltiorrhiza. (B) Schematic presentations for image process, feature extraction, and analysis for agronomic traits using WinRHIZO and RhizoVision. (C) Schematic presentations for anatomy traits analyses using sections of S. miltiorrhiza roots. (D) Schematic presentations for modeling of root architecture structure using landmark method. (E) Schematic presentations for metabolic profiles on the section of S. miltiorrhiza roots using mass spectrometry imaging.

Fig. 2.

Phenotyping of S. miltiorrhiza root based on scanning images of the whole root system. (A) The workflow used for root phenotyping and data analysis. High-throughput imaging data from the WinRHIZO scanning system were imported and processed using WinRHIZO and RhizoVision. The extracted phenotypic traits were further evaluated for biomass estimation and RSA study. (B) The distribution of parameters related to the biomass of S. miltiorrhiza roots. (C) Linear regression analysis between real biomass (Fresh weight and Dry weight) and extracted parameters (Total Length, Total Surface, and Total Volume). (D) K-means clustering of S. miltiorrhiza roots samples (n = 92) according to the diameter ranges of each root branch. The root model represented the typical root structure among each cluster. The bar represented the size of each root model. (E) The importance evaluation of phenotypic traits according to the K-means clustering using the random forest model.

Fig. 3.

RSA analyses of S. miltiorrhiza roots by the landmark method. (A) Landmark placement for S. miltiorrhiza root. We defined a set of 9 landmarks covering the overall outline of the root system and primary roots. (B) Samples of S. miltiorrhiza root were grouped into 10 clades according to the landmark placement. The branch of each group was labeled in different colors. The clades were then clustered into 3 categories by PCA (C). (D) The importance evaluation of phenotypic traits according to the landmark clustering (10 clades) using the random forest model.

Fig. 4.

Metabolic profiling of pharmaceutic metabolites in S. miltiorrhiza roots and their correlation with phenotypic traits. (A) Mass spectrometry imaging by MALDI-MS showed the spatial distribution of 2 classes of effective metabolites on a section of S. miltiorrhiza root with a spatial resolution of 75 μm. The ion strength is color-coded (white = maximal signal and black = minimal signal) and normalized. (B) Content of phenolic acids and tanshinones in different root tissues measured by HPLC-MS/MS. X, xylem; W, phloem and cambium layers; P, periderm, F, the whole root segment. (C) ML represented the most variable metabolic traits according to K-means clustering shown in Fig. 2E. The top 3 parameters were labeled. Linear regression analysis showed that the production of LAB (D) was significantly correlated to the selected phenotypic trait (Total Surface), while tanshinone IIA (E) was small but significantly correlated to the selected phenotypic trait (Total Surface).