Genome-Wide Network Analysis of Above- and Below-Ground Co-growth in Populus euphratica

Abstract

Tree growth is the consequence of developmental interactions between above- and below-ground compartments. However, a comprehensive view of the genetic architecture of growth as a cohesive whole is poorly understood. We propose a systems biology approach for mapping growth trajectories in genome-wide association studies viewing growth as a complex (phenotypic) system in which above- and below-ground components (or traits) interact with each other to mediate systems behavior. We further assume that trait-trait interactions are controlled by a genetic system composed of many different interactive genes and integrate the Lotka-Volterra predator-prey model to dissect phenotypic and genetic systems into pleiotropic and epistatic interaction components by which the detailed genetic mechanism of above- and below-ground co-growth can be charted. We apply the approach to analyze linkage mapping data of Populus euphratica, which is the only tree species that can grow in the desert, and characterize several loci that govern how above- and below-ground growth is cooperated or competed over development. We reconstruct multilayer and multiplex genetic interactome networks for the developmental trajectories of each trait and their developmental covariation. Many significant loci and epistatic effects detected can be annotated to candidate genes for growth and developmental processes. The results from our model may potentially be useful for marker-assisted selection and genetic editing in applied tree breeding programs. The model provides a general tool to characterize a complete picture of pleiotropic and epistatic genetic architecture in growth traits in forest trees and any other organisms.

Graphical abstract

Fig. 1.

The workflow of genome-wide multilayer networks that mediate complex dynamic traits. (A) Collection of time-series phenotypic data and alleles of stem height, taproot length, lateral root length, and average lateral root length in Populus euphratica. (B) Construct and fit the multi-dimensional interactive model (MDIM). (C) Multivariate systems mapping to identify key QTLs regulating the growth of multiple traits and their interactions. (D) Obtain the optimal number of modules for functional clustering according to the BIC criterion. (E) Build a network (bidirectional, signed, and weighted) among modules (nodes) and subdivide the modules further until a node represents an SNP. (F) Simulation experiments to verify the validity of our model.

Fig. 2.

Growth curves of (A) stem length (ST), (B) taproot length (TAP), (C) lateral root length (LA), and (D) average lateral root length (ALA) of Populus euphratica. Phenotypic observations of ST, TAP, LA, and ALA of all seedlings are indicated by thin solid lines, corresponding to blue, red, green, and gray, respectively. Average growth values (black scatter) are fitted with MDIM (thick solid lines). The fitting curve of each trait is composed of independent growth (thick broken lines) and dependent growth on the other 3 traits (thick dot lines).

Fig. 3.

Growth curve fitting and coefficient of variation analysis. (A) Average fitting curves and (B) growth rate curves of stem length (blue), taproot length (red), lateral root length (green), and average lateral root length (gray). (C) Combination of scatterplot and boxplot of residual values. (D) Curves of CV values.

Fig. 4.

Analysis of identified QTLs. (A) Manhattan plot of LR values for 19 linkage groups, and the red horizontal line represents the threshold. (B) includes cellular component (CC), molecular function (MF), and biological process (BP). Bar chart (C) shows the percentage of key QTLs identified by systems mapping and successfully annotated QTLs in each linkage group.

Fig. 5.

Schematic diagram of the multilayer interactive network model. In the first layer, the whole genome is a large-scale network composed of modules detected by functional clustering. Ellipse shadows with different colors stand for modules that include SNPs with similar dynamic patterns. There are activation (red link) and inhibition (black link) between modules. In the second layer, each module in the first layer is divided into sub-modules with connections. As an example, the pink module contains 3 sub-modules with links. In the last layer, each sub-module representing a small network community contains closely linked variables that are less closely linked with variables from other modules. A small circle within the sub-module represents a minimum unit, i.e., an SNP.

Fig. 6.

Modules and networks of genetic effects. (A) Genetic effect curves of SNPs in 11 representative modules on stem length (blue line), taproot length (red line), lateral root length (green line), and average lateral root length (orange line) chosen from 105 modules; BIC curve shows that the optimal number of modules is 105. (B) Genetic networks of 105 modules for 4 above- and below-ground traits; there are activation (pink arrowed lines) and inhibition (blue arrowed lines) among modules, with the thickness proportional to the intensity of regulation. (C) The distribution of the number of outgoing modules (upper part) and incoming modules (lower part) of the networks for 4 traits. The x-axis is obtained by arranging 105 modules in descending order according to the number of their outgoing modules.

Fig. 7.

Modules and networks of genetic effects of M24. (A) Genetic effect networks of 4 above- and below-ground traits. Each node represents an SNP or QTL. The directed arrowed lines between nodes represent the interaction relationship, the green one represents the inhibitory effect, the red one represents the promoting effect, and the thickness of the edge is proportional to the strength of the interaction. The brown node represents QTL. (B) The distribution of the number of outgoing modules and incoming modules of 4 traits across SNPs is given.