A major workflow in Systems Biology is an interlocking circle between experimental and theoretical work . Experimentalists perform high-throughput experiments in order to monitor the response of a biological system to an external stimulus. These data is then used to construct spatio-temporal models from which reasonable hypotheses are generated. These hypotheses are experimentally verified or falsified. Using the results of these experiments, scientists are able to refine the model and thus generate new knowledge .
One way of describing biological systems are networks. Networks are graphical representations, where the nodes represent the objects of interest and edges represent relations between these objects . Network models help to explain, understand and describe the functioning of a cell . In many cases we do not know the underlying interaction networks within the system of interest. Network inference aims at the deduction of these networks utilizing high-throughput data and prior knowledge. The inference of gene regulatory networks consists of three parts: the identification of potential regulators, the prediction of target genes, and the inference of the mode of interaction (e.g. activation or repression). A number of approaches are established to perform this task, such as setting up Bayesian Networks , information theoretical approaches [6–8], regression based inference [9–11], and differential equation models [12–17]. A number of studies successfully applied these methods for different biological purposes, e.g. modelling of immune diseases [10, 13], full genomic models of Escherichia coli  and Saccharomyces cerevisiae , and models of pathogenic fungi . It has been shown that the integration of different data sources improves the reverse engineering approach [10, 19–21].Since different data sources might be contradictory, it is advantageous to softly integrate them during the modelling procedure. That means, proposed interactions can be scored by the confidence of the prior knowledge source and might be removed if they contradict too much to the measured data. A recent study shows how the Systems Biology circle supports network inference . Due to the large amount of available data and knowledge E. coli is best suited as model organism for network inference. However, this task is more difficult for pathogenic fungi by virtue of the small amount of data and small number of known interactions.
Aspergillus fumigatus is an airborne saprophytic fungus . Humans constantly inhale numerous conidia of A. fumigatus, which are usually eliminated by the immune system. However, in immunocompromised individuals the fungus can cause life-threatening infections . In fact, the number of infections has been dramatically increased due to the growing number of immunocompromised individuals [24–26].
The human host evolved a number of strategies to prevent microbial infection. One important strategy is to keep iron away from the pathogen . Iron is an essential metal required as a cofactor for several proteins, as well as for a number of biochemical processes. However, within the human host, iron is bound to proteins such as haemoglobin, ferritin, transferrin, and lactoferrin. Consequently, there is almost no free iron available . Thus, the acquisition of iron is an important virulence attribute of most pathogens. During co-evolution, A. fumigatus has developed a number of efficient iron acquisition pathways: 1) reductive iron uptake, 2) uptake via siderophores, and 3) low-affinity uptake (for a more detailed description see ). Since excess of iron is toxic for a cell, iron homeostasis needs to be tightly regulated in A. fumigatus. The knowledge about the molecular interactions underlying these regulations is still fragmentary. The transcription factors SreA and HapX have been identified as a counter pair [30–32]. Under iron replete conditions, SreA is activated and represses iron uptake. Under these conditions, SreA also represses hapX transcription. Since HapX is a repressor of iron consumption pathways, SreA indirectly activates iron consumption. Moreover, HapX also acts as an activator of iron acquisition. A number of target genes are known for both regulators, however we are still far from a complete understanding of iron homeostasis in A. fumigatus.
Recently, we proposed a model predicting regulatory interactions for iron uptake of another fungal pathogen, Candida albicans, when the fungus is adhering to and invading into human epithelial cells . The model is based on time series expression data during experimental infection of reconstituted human oral epithelium. The usefulness of these data lies in the fact that it re-samples important parts of a real infection scenario. On the other hand, in the previous modelling approach a number of environmental parameters are not constant during infection, such as pH and nutrient availability. This may have caused side effects and made it difficult to decide whether the proposed interactions are purely based on changes in iron availability or other environmental parameters. Such environmental variations finally hamper experimental verifications of the proposed interactions. The use of in vitro time series expression data after a change from iron depleted to iron replete conditions will help to decide which interactions C. albicans uses to regulate iron homeostasis. For A. fumigatus, such time series expression data is already available and utilized in this study.
In the present work, we propose the first computational model of the regulation of iron homeostasis genes in A. fumigatus using high-throughput gene expression time series data after a shift from iron starvation to iron replete conditions . It is based on a set of linear differential equations and utilizes selection criteria such as sparseness and robustness [17, 21, 33]. Since the soft integration of prior knowledge has been shown to improve the reliability of the predicted networks [10, 19–21], our modelling approach softly integrates three kinds of prior knowledge: Northern blot analysis under limited iron [31, 32], microarray expression analysis of transcription factor knock-out mutants [31, 32], as well as the occurence of transcription factor binding motifs analysis in regulatory regions of genes [31, 34–36]. The inferred model predicts new transcription factor to target gene interactions. A recent study utilizes Northern blots and experimentally verifies two of these interactions , while another predicted interaction is falsified and one remains unevaluated. Using the results of the recent experiments as additional prior knowledge, we are able to refine our model. The final network model predicts a number of SrbA targets. To study, whether or not the transcriptional regulator directly binds to these target genes, we performed motif searching that lead to the identification of potential SrbA binding sites in the promoters of the predicted target genes. Indeed, wet-lab experiments demonstrate high-affinity binding capacity of SrbA to the promoters of hapX, hemA and srbA.