Systems biology of plant molecular networks: from networks to models

Developmental processes are controlled by regulatory networks (GRNs), which are tightly coordinated networks of transcription factors (TFs) that activate and repress gene expression within a spatial and temporal context. In Arabidopsis thaliana, the key components and network structures of the GRNs controlling major plant reproduction processes, such as floral transition and floral organ identity specification, have been comprehensively unveiled. This thanks to advances in ‘omics’ technologies combined with genetic approaches. Yet, because of the multidimensional nature of the data and because of the complexity of the regulatory mechanisms, there is a clear need to analyse these data in such a way that we can understand how TFs control complex traits. The use of mathematical modelling facilitates the representation of the dynamics of a GRN and enables better insight into GRN complexity; while multidimensional data analysis enables the identification of properties that connect different layers from genotype-to-phenotype. Mathematical modelling and multidimensional data analysis are both parts of a systems biology approach, and this thesis presents the application of both types of systems biology approaches to flowering GRNs. Chapter 1 comprehensively reviews advances in understanding of GRNs underlying plant reproduction processes, as well as mathematical models and multidimensional data analysis approaches to study plant systems biology. As discussed in Chapter 1, an important aspect of understanding these GRNs is how perturbations in one part of the network are transmitted to other parts, and ultimately how this results in changes in phenotype. Given the complexity of recent versions of Arabidopsis GRNs - which involves highly-connected, non-linear networks of TFs, microRNAs, movable factors, hormones and chromatin modifying proteins - it is not possible to predict the effect of gene perturbations on e.g. flowering time in an intuitive way by just looking at the network structure. Therefore, mathematical modelling plays an important role in providing a quantitative understanding of GRNs. In addition, aspects of multidimensional data analysis for understanding GRNs underlying plant reproduction are also discussed in the first Chapter. This includes not only the integration of experimental data, e.g. transcriptomics with protein-DNA binding profiling, but also the integration of different types of networks identified by ‘omics’ approaches, e.g. protein-protein interaction networks and gene regulatory networks. Chapter 2 describes a mathematical model for representing the dynamics of key genes in the GRN of flowering time control. We modelled with ordinary differential equations (ODEs) the physical interactions and regulatory relationships of a set of core genes controlling Arabidopsis flowering time in order to quantitatively analyse the relationship between their expression levels and the flowering time response. We considered a core GRN composed of eight TFs: SHORT VEGETATIVE PHASE (SVP), FLOWERING LOCUS C (FLC), AGAMOUS-LIKE 24 (AGL24), SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1 (SOC1), APETALA1 (AP1), FLOWERING LOCUS T (FT), LEAFY (LFY) and FD. The connections and interactions amongst these components are justified based on experimental data, and the model is parameterised by fitting the equations to quantitative data on gene expression and flowering time. Then the model is validated with transcript data from a range of mutants. We verify that the model is able to describe some quantitative patterns seen in expression data under genetic perturbations, which supported the credibility of the model and its dynamic properties. The proposed model is able to predict the flowering time by assessing changes in the expression of the orchestrator of floral transition AP1. Overall, the work presents a framework, which allows addressing how different quantitative inputs are combined into a single quantitative output, i.e. the timing of flowering. The model allowed studying the established genetic regulations, and we discuss in Chapter 5 the steps towards using the proposed framework to zoom in and obtain new insides about the molecular mechanisms underlying the regulations. Systems biology does not only involve the use of dynamic modelling but also the development of approaches for multidimensional data analysis that are able to integrate multiple levels of systems organization. In Chapter 3, we aimed at comprehensively identifying and characterizing cis-regulatory mutations that have an effect on the GRN of flowering time control. By using ChIP-seq data and information about known DNA binding motifs of TFs involved in plant reproduction, we identified single-nucleotide polymorphisms (SNPs) that are highly discriminative in the classification of the flowering time phenotypes. Often, SNPs that overlap the position of experimentally determined binding sites (e.g. by ChIP-seq), are considered putative regulatory SNPs. We showed that regulatory SNPs are difficult to pinpoint among the sea of polymorphisms localized within binding sites determined by ChIP-seq studies. To overcome this, we narrowed the resolution by focusing on the subset of SNPs that are located within ChIP-seq peaks but that are also part of known regulatory motifs. These SNPs were used as input in a classification algorithm that could predict flowering time of Arabidopsis accessions relative to Col-0. Our strategy is able to identify SNPs that have a biological link with changes in flowering time. We then surveyed the literature to formulate hypothesis that explain the regulatory mechanism underlying the difference in phenotype conferred by a SNP. Examples include SNPs that disrupt the flowering time gene FT; in which the mutation presumably disrupts the binding region of SVP. In Chapter 5 we discuss the steps towards extending our approach to obtain a more comprehensive survey of variants that have an effect on the flowering time control. In Chapter 4, we propose a method for genome-wide prediction of protein-protein interaction (PPI) sites form the Arabidopsis interactome. Our method, named SLIDERbio, uses features encoded in the sequence of proteins and their interactions to predict PPI sites. More specifically, our method mines PPI networks to find over-represented sequence motifs in pairs of interacting proteins. In addition, the inter-species conservation of these over-represented motifs, as well as their predicted surface accessibility, are take into account to compute the likelihood of these motifs being located in a PPI site. Our results suggested that motifs overrepresented in pairs of interacting proteins that are conserved across orthologs and that have high predicted surface accessibility, are in general good putative interaction sites. We applied our method to obtain interactome-wide predictions for Arabidopsis proteins. The results were explored to formulate testable hypothesis for the molecular mechanisms underlying effects of spontaneous or induced mutagenesis on e.g. ZEITLUPE, CXIP1 and SHY2 (proteins relevant for flowering time). In addition, we showed that the binding sites are under stronger selective pressure than the overall protein sequence, and that this may be used to link sequence variability to functional divergence. Finally, Chapter 5 concludes this thesis and describes future perspectives in systems biology applied to the study of GRNs underlying plant reproduction processes. Two key directions are often followed in systems biology: 1) compiling systems-wide snapshots in which the relationships and interactions between the molecules of a system are comprehensively represented; and 2) generating accurate experimental data that can be used as input for the modelling concepts and techniques or multi-dimensional data analysis. Highlighted in Chapter 5 are the limitations in key steps within the systems biology framework applied to GRN studies. In addition, I discussed improvements and extensions that we envision for our model related to the GRN underlying the control of flowering time. Future steps for multi-dimensional data analysis are also discussed. To sum up, I discussed how to connect the different technologies developed in this thesis towards understanding the interplay between the roles of the genes, developmental stages and environmental conditions.