Elucidating regulatory mechanisms downstream of a signaling pathway using informative experiments

Ewa Szczurek^1,2,3, Irit Gat-Viks^1,a, Jerzy Tiuryn³ & Martin Vingron¹

^aPresent address: Broad Institute of MIT and Harvard, 7 Cambridge Center, Cambridge, MA 02142, USA

Article highlights

Synopsis

Signaling cascades are triggered by extracellular stimulation and propagate the signal downstream to the transcriptional machinery. The process is mediated by regulators in the signaling network that may act directly or indirectly as repressors or activators of certain target genes, so that the genes are up- or downregulated depending on the activity state of the regulators in a given condition. Experimental studies in which a specific signaling cascade is perturbed to investigate its downstream regulation mechanisms (rather than global mapping of cellular transcription) became common in the recent years (e.g., Roberts et al, 2000; Yoshimoto et al, 2002; O'Rourke and Herskowitz, 2004). In these studies, the signaling network is manipulated either by external stimulation or by perturbation of its molecules, by knockout or overexpression. Each such experiment alters the activation status of the regulators in the network, and consequently, the readout on their downstream target genes. It is impractical to perform all possible manipulations of the signaling pathway under study, and therefore the researcher must prioritize and choose only the most informative manipulations. In this paper, we present a new systems biology framework guiding the choice of experiments in research investigating transcription regulation downstream of a particular signaling network.

The framework consists of three basic components: logical model of the signaling network under study, experimental design algorithm called MEED, and identification of regulator–target relationships, performed by an expansion procedure. The logical model formalizes qualitative knowledge about the signaling network, representing biological components, such as signaling molecules, environmental stimulations and transcription factors, as well as the logical relations among them (Gat-Viks et al, 2004). Most importantly, the model is predictive: For a given network manipulation, the model pre-calculates the activation states of the regulators in the network, as well as the predicted expression profiles of all putative targets of the pathway.

The MEED algorithm aims to select experiments from a set of candidates optimizing two objectives: (i) to minimize the number of selected experiments and (ii) to maximize diversity between expression profiles of genes regulated through different mechanisms. The second condition aims to avoid an ambiguous situation in which two genes with distinct regulatory mechanisms attain the same expression profile under the selected experiments. Only in the case in which the two genes have two distinct expression profiles, it is possible to distinguish their regulatory mechanisms. The algorithm utilizes the model predictions about all possible predicted expression profiles of the potential target genes. With the model predictions, MEED is able to design a set of experiments without access to any additional data. MEED is the first experimental design algorithm that can be applied to the task of target identification based only on qualitative knowledge, when no prior experimental data (gene expression or transcription factor binding) is available. Moreover, unlike extant approaches, our algorithm considers potential dependencies between the suggested experiments. With this innovative feature, MEED can design a set of informative, non-redundant experiments that can be efficiently carried out in parallel.

For each target gene, the expansion procedure aims to identify the target's regulator in the network and the way the regulation takes place. To figure out these regulatory relationships, the procedure compares the regulators' activity states with gene expression response to different experimental manipulations. The gene responses are assessed using high-throughput gene expression measurements, whereas the regulators' activation states are predicted by the logical model. The expansion procedure relies on the quality and information content of the experiments it uses to compare the measured and model-predicted gene responses.

The three components of our framework form an iterative pipeline. The researcher's prior knowledge about the studied pathway should be formalized in a logical model and provided as input to MEED, together with a list of candidate experiments and a set of regulators of interest. Given the model predictions, the algorithm selects experiments optimized for unambiguous identification of regulator–target relationships downstream of the model. MEED instructs the researcher under which environmental conditions and with what perturbations the experiments should be conducted. The suggested experiments are prioritized such that the researcher can choose to carry out only the most informative ones. Next, gene expression is measured in a lab under the suggested experimental manipulations. This data, together with model predictions, is then fed to the expansion procedure to produce hypotheses about regulator–target relationships. Results of the expansion can be iteratively improved by applying MEED to choose experiments to complement those that were already carried out.

The framework was utilized to identify gene regulatory modules downstream of a network of three interconnected yeast MAPK pathways. Using experiments chosen by MEED, we applied the expansion procedure and identified regulatory modules comprising groups of genes co-regulated by molecules in the pathway through a specific regulatory mechanism. We iterated experimental design to propose additional experiments for resolving the ambiguity remained after the expansion step. In comparison with other approaches, the experiments suggested by MEED make it possible to draw less ambiguous conclusions about transcriptional regulation. Moreover, our comparative analysis shows the importance of considering dependencies between experiments as part of the experimental design process.

MORE ARTICLES LIKE THIS

These links to content published by NPG are automatically generated.