A systematic strategy that relies on both in silico modeling and experimental data was developed to help predict the origins of drug actions. Our systems biology strategy may also provide a rationale for explaining differential and common effects of drugs and drug classes. The prediction method uses information derived from the chemical structure of the drugs combined with experimental omics data obtained from short-term efficacy studies. Overall, we demonstrated notable consistencies between the computationally predicted effects and the wet-lab biochemical effects. The results suggested that the systems biology-based approach may facilitate identification of on-target and off-targets effects of a new drug which can be determined in the early stages of the drug development cycle. As importantly, the strategy may improve understanding of drug efficacies and aid in predicting safety leading to reduced costs of drug development and drug attrition rates. This approach may have important applications and implications for preclinical research and for development of novel therapeutic strategies including drug combinations.
Prototype cardiovascular drugs were used to evaluate our strategy. Differential effects on lipid metabolism and inflammation as obtained from in silico predictions were experimentally confirmed. A key step of the strategy is the use of computational predictions using structural data of the parent compound, compounds in the same chemical class, and their metabolites. The power of this approach is that variation in the structure of compounds belonging to a specific class allows a greater chance of finding other protein-protein interactions that are relevant for understanding primary and secondary effects of the drug. The MetaDrug database is very large and contains over 700,000 chemical structures and approximately 500,000 protein-compound interactions (covering about 4,500 protein targets), increasing the number of chemically-similar compounds and resulting in an increased power of the analyses. An added strength of this study was the ability to computationally expand from primary targets to their closest network interactions. Expanding the network increased the number of pathways and genes that could be compared thereby increasing the ability to predict effects on biological processes involved in the phenotype or disease process.
The hypotheses generated by these prediction tools and database were experimentally validated by (i) in vitro and in vivo techniques that quantified plasma levels of lipids and inflammation markers, (ii) by analyzing activity of key transcription factors in livers, and (iii) analysis of atherosclerotic plaque size in the aorta. It was predicted that the biological pathways maps that RSV, T09 and FF would affect would hardly overlap and that similarities between these drugs could mainly be expected within the maps of lipid metabolism and inflammation. Among the processes predicted to be affected by all three drugs were IL-1 signaling and MIF signaling and among the 21 common targets predicted were PPARγ, C/EBP, SMAD3, p53, SP1, Rb protein, COX-2, ERK1/2 and JNK. In silico network analyses also predicted that FF would affect the most network processes and that FF would differ in its impact on inflammation (with RSV and T09 being more similar). The experimental validation using the ApoE*3Leiden mice treated with the drugs specifically focused on inflammatory networks. The data confirmed the predictions: all three drugs affected inflammatory processes controlled by IL-1 and MIF and mediated by C/EBP, SP1, ERK1/2 and JNK. Also, the drugs showed a specific pattern of anti-inflammatory action with RSV and T09 sharing more similarities (for instance activation of central transcription factors C/EBPβ and p65-NFκB). In accordance with the model prediction, FF differed markedly from RSV and T09 in its overall effects on inflammation. While RSV and T09 mainly affected acute inflammatory response processes, FF was more effective in controlling chronic inflammation processes. A notable prediction was that FF would quench IL-6 signaling and related downstream effects relative to RSV and T09, which was confirmed at the protein level by transcription factor activity analysis (reduced STAT3 activity) and by the observed FF-specific reduction of circulating fibrinogen, an IL-6-inducible acute phase protein and chronic inflammation marker. Collectively, comparative genome-wide pathway mapping showed that the biological effects of the drugs were realized largely via different pathways and mechanisms suggesting complementarities.
Transcriptome data was used to predict physiological effects relevant for the vasculature and, to a lesser extent, for the liver itself. For example, FF (but not RSV or T09) was predicted to have an effect on processes important for leukocyte activation, migration and recruitment, all of which are crucial processes in early atherogenesis. Indeed, FF had the strongest effect on early atherosclerotic lesion development under the applied experimental conditions. Thus, while all three drugs were expected to be anti-atherosclerotic, FF is predicted to quench immune and inflammatory processes which play an important role in early lesion formation. This effect distinguishes FF from RSV and T09 and suggested that FF would be more potent than the other two drugs. Indeed, FF had the strongest reducing effect on early atherosclerosis. Extending the observations for processes and drug effects to other tissues will require additional confirmatory experiments. Although an increasing number of published studies use transcriptomic data from circulating cells to extrapolate to biological effects in other tissues and the effects of drugs , insufficient data exists to estimate the value and limitations of these strategies.
An increasing number of studies have shown that several cardiovascular drugs originally designed to lower plasma lipid levels also have beneficial anti-inflammatory effects, specifically the down-regulation of major inflammatory markers (TNFα, interleukin-1β, fibrinogen, SAA and CRP) and several key inflammatory transcriptional regulators (NF-κB). These effects were described by us and others for hypolipidemic drugs of different classes: statins, fibrates and LXR agonists [15, 29–33]. However, the pathways and mechanisms that explain these anti-inflammatory effects remained largely unknown including whether these drugs act on the same or different pathways. Our data showed differential activities on inflammatory processes with signaling pathways and specifically via the key regulators including interferon-gamma, TGFβ, IL-1, TNFα, MIF and IL-6. These experimental data indicated that the profound inflammation quenching effect of FF may be through its effect on IL-6 signaling, a result consistent with the global suppression of IL-6-regulated genes by FF  and its negative (PPAR-alpha-dependent) effect on the IL-6 target gene fibrinogen . However, the targets of FF and T09 are nuclear hormone receptors which are expressed in a cell- and tissue-specific manner and our observations are based on analyses of liver after chronic exposure to these drugs. Whether similar effects apply to other tissues and whether these affects also persist under conditions of chronic drug exposure remains to be experimentally tested. Many current systems biology-based strategies rely upon data from one organ that is composed of multiple types of cells, a distinct limitation of existing tissue isolation and analyses technologies.
Functional systems and pathway analyses methods capable of analyzing complex, multi-gene biological phenotypes are rapidly developing and are likely to help in understanding the mechanisms of drug effects. A structured "knowledge base" consisting of protein-protein interactions, pathways and processes assembled in ontologies  is required for such analyses. The data used to generate the lists of molecules (genes, proteins, metabolites) for prediction and pathway analysis may, however, be derived from experiments in different species, methods, and strategies. As shown here, the ability to correctly predict experimental results indicated the utility and potency of systems biology strategies in general and for translating results from laboratory animals models to the human. Nevertheless, these strategies are currently a method for hypothesis generation and the results, however promising, have to be considered "predictions." The three compounds analyzed are well known drugs which have a lot of associated publications in the literature and consequently in the MetaDrug database. The predictions made herein are, however, only partially based upon existing literature connections. On basis of the chemical structure of the compounds, our method also extrapolates to possible metabolites (formed after liver passage) and their respective targets - this portion of the prediction process is solely based on the chemical structure of the drugs and can be viewed as true predictions.
Developing new drugs is a tedious and expensive undertaking. Despite improvements in rational drug design and high throughput screening methods, the number of novel, single-target drugs fell greatly behind expectations during the past decade. In addition, the treatment of complex diseases involving multiple genes and risk factors remains a pressing medical need. The effects of drugs on known or unsuspected targets present both opportunities and challenges for modern drug discovery. Developing high-efficacy drugs that alter the activity of multiple targets or repositioning existing drugs to treat different diseases highlight the possibilities of a systems biology approach. However, off-target effects may result in adverse drug reactions that account for around one-third of drug failures during development and may contribute to idiosyncratic drug-induced damage to tissues. Reliable and reproducible strategies and models for predicting efficacy and safety, particularly in being able to identify the direct and indirect targets early in the drug development process are greatly needed. Such strategies are increasingly relevant for the development of successful combination therapies for patients suffering from complex, multifactorial cardiometabolic pathologies. Examples include patients treated with one or more drugs such as lipid-lowering and/or hypotensive drug therapies. This report provides an example of and extends the scope of systems biology approaches for drug discovery.