The phenotypic diversity of Schisandra sphenanthera fruit and SVR model for phenotype forecasting

Abstract

Schisandra sphenanthera Rehd. et Wils. is an endangered traditional Chinese medical plant whose fruits can be used for medicinal purposes, and is mainly distributed in the Qinling Mountains. Presently, little is known about the phenotypic diversity in natural populations of S. sphenanthera, and its fitness and evolutionary potential to the environmental factors, which is important for its conservation. In this study, we have made an in-depth research of the degree of phenotypic differentiation between and within populations, phenotypic diversity, environment factors on phenotypic differentiation, and the prediction of fruit phenotypes under different environments. The phenotypic diversity of S. sphenanthera fruits on a large scale (9 provinces, 25 counties) is evaluated by means of cluster analysis, principle component analysis, statistics of phenotypic traits diversity index and differentiation coefficient. Our data report a high degree of variation between the 25 populations for most morphological traits. The average phenotypic variation between populations accounts for 27.11%, and the average phenotypic variation within populations accounts for 72.89%. 25 populations are divided into four categories based on twelve fruit phenotypic traits of S. sphenanthera. Five counties in Shaanxi Province are grouped together (cluster I) where the highest single fruit weight and 100-berries fresh weight, which are most closely related to fruit weight. Correlation analysis is used to evaluate the relationship between environmental factors and fruit phenotypes, and path analysis is used to analyse the effect of environmental factors on fruit phenotypes. There are many significant correlations between phenotype and environmental factors. Some factors have direct effects on phenotype, while others are indirect effect factors, for example, bio1 has direct effects on longitudinal and horizontal diameter of fruit and 100-berries weight (including fresh and dry weight), and has indirect effect on fruit pedicels length. Topographic and soil factors mainly affect phenotypes about fruit weight, and phenotypes of fruit shape are affected by topographic and climate factors. Our study propose a data-driven machine learning (ML) method, namely support vector regression (SVR), based on the results of path analysis, which can predict the phenotype of S. sphenanthera in different environments with high prediction accuracy. S. sphenanthera in five counties of Shaanxi Province (cluster I) can serve as an important source of genetic material for wild upbringing towards the development of new breed with high yield in this Chinese herb. This study also finds the environmental factors have significant effect on fruit weight-related phenotypes. The results of this study can provide data support for artificial cultivation to improve fruit yield.