Systems-ADME/Tox: Resources and Network Approaches
Total Page:16
File Type:pdf, Size:1020Kb
Journal of Pharmacological and Toxicological Methods 53 (2006) 38 – 66 www.elsevier.com/locate/jpharmtox Appraisal of state-of-the art Systems-ADME/Tox: Resources and network approaches Sean Ekins * GeneGo, 500 Renaissance Drive, Suite 106, St. Joseph, MI 49085, USA School of Pharmacy Department of Pharmaceutical Sciences, University of Maryland, USA Received 23 May 2005; accepted 23 May 2005 Abstract The increasing cost of drug development is partially due to our failure to identify undesirable compounds at an early enough stage of development. The application of higher throughput screening methods have resulted in the generation of very large datasets from cells in vitro or from in vivo experiments following the treatment with drugs or known toxins. In recent years the development of systems biology, databases and pathway software has enabled the analysis of the high-throughput data in the context of the whole cell. One of the latest technology paradigms to be applied alongside the existing in vitro and computational models for absorption, distribution, metabolism, excretion and toxicology (ADME/Tox) involves the integration of complex multidimensional datasets, termed toxicogenomics. The goal is to provide a more complete understanding of the effects a molecule might have on the entire biological system. However, due to the sheer complexity of this data it may be necessary to apply one or more different types of computational approaches that have as yet not been fully utilized in this field. The present review describes the data generated currently and introduces computational approaches as a component of ADME/Tox. These methods include network algorithms and manually curated databases of interactions that have been separately classified under systems biology methods. The integration of these disparate tools will result in systems-ADME/Tox and it is important to understand exactly what data resources and technologies are available and applicable. Examples of networks derived with important drug transporters and drug metabolizing enzymes are provided to demonstrate the network technologies. D 2005 Elsevier Inc. All rights reserved. Keywords: Algorithms; Human; Microarray; Mouse; Networks; Rat; Software; Toxicogenomics; Toxicoproteomics 1. Introduction in the industry as they have been evaluated primarily for toxicology and metabolism assessment (Waring et al., Metabolism and safety assessment have witnessed some 2003, 2002, 2001) with some considerable focus on growth in the number of new technologies and methods hepatotoxicity (Harris, Dial, & Casciano, 2004; Hartley that have been introduced within the last decade. However, et al., 2004; Heijne et al., 2004; Huang et al., 2004; according to a recent FDA white paper there is still Liguori et al., 2005; Ulrich, Rockett, Gibson, & Pettit, considerable scope for additional new methods (FDA, 2004). For example, searching PubMed for publications in 2004). For example, recently various reports have the last 5 years with the keywords Fmicroarray and described new software and methods for metabolism toxicology_ or Ftoxicogenomics_, indicates that the accu- prediction (Balakin et al., 2004a, 2004b; Borodina et al., mulation of papers describing the latter is doubling every 2003; Borodina et al., 2004; Boyer & Zamora, 2002; year (Fig. 1) which perhaps is mirrored by the application Korolev et al., 2003). Simultaneously the use of high in the pharmaceutical industry for predictive toxicology throughput (HT) methods for genomics, proteomics and (Suter, Babiss, & Wheeldon, 2004). metabonomics have taken off in terms of their acceptance To date toxicogenomics experiments have been carried out under non-standardized conditions. Most of the studies * Corresponding author. Tel.: +1 269 930 0974; fax: +1 269 983 7654. have been conducted with rats, less often mice, using E-mail addresses: [email protected], [email protected]. multiple different microarray formats and statistical proce- 1056-8719/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.vascn.2005.05.005 S. Ekins / Journal of Pharmacological and Toxicological Methods 53 (2006) 38–66 39 100 The perturbing effect of a molecule on the complete 80 biological system can be observed across all metabolic and signaling pathways or networks and can provide some 60 limited insight into the binding to multiple proteins or effects on gene expression simultaneously. This requires the 40 PubMed collection of high-throughput data, including global gene expression, protein content, metabolic profiles for the same 20 samples as well as individual genetic, clinical and pheno- Number of publications in 0 typic data. However there are difficulties with such an 2000 2001 2002 2003 2004 2005 approach as there are likely to be differences between the Year Fstimulus to effect_ durations for all the gene–protein relationships (Nicholson, Holmes, Lindon, & Wilson, Fig. 1. Annual frequency of articles appearing with the words ‘‘toxicoge- nomics’’ (squares) or ‘‘microarray and toxicology’’ (diamonds). 2004). We can now use either the growing number of academic or commercially available pathway database and network dures (Tables 1-4). There have been relatively few cross- building tools with expression data. These enable the platform toxicogenomics studies under controlled condi- connection of interacting, differentially expressed genes as tions (Thompson et al., 2004). Despite the suggested poor networks (Barabasi & Oltvai, 2004; Hanisch, Zien, Zimmer, compatibility between the different array types, this latter & Lengauer, 2002; Ideker, Ozier, Schwikowski, & Siegal, study demonstrated a high (90%) consistency between the 2002; Ideker et al., 2001; Segal et al., 2003a; Segal, Wang, & expression of the genes that were shared between the Koller, 2003b; Spirin & Mirny, 2003; Tornow & Mewes, platforms. The development of methods to visualize such 2003) as well as allowing the reverse engineering of complex expression data has also expanded beyond the functional connections (Somogyi, Fuhrman, & Wen, 2001). widely used clustering methods (Eisen, Spellman, Brown, & The use of such network visualizations suggests an Botstein, 1998). With the outcome of microarray analysis organized modularity in complex systems (Han et al., being dependent on the widely used statistical procedures 2004) which has also been applied to interpret the applied to derive those genes that are significantly differ- connectivity of small molecules and their interaction with entially expressed (Butte, 2002), newer approaches that do proteins in the subfield of chemogenomics (Bredel & Jacoby, not necessarily require data clustering may be an advantage. 2004; Csermely, Agoston, & Pongor, 2005; Parsons et al., As rat and mouse are the most widely used in vivo toxicity 2004; Sharom, Bellows, & Tyers, 2004). The parallel models it is assumed that acute and chronic toxicity shown development of HT methods, databases, ADME/Tox model- in animals largely coincides with human toxicity. Therefore ing and systems modeling is ongoing (Ekins, Nikolsky, & differential expression patterns in animal models are also Nikolskaya, 2005e). The present review is therefore timely assumed to be predictive of the end point toxic response in as it discusses some of the data resources, limitations and human. This is not always the case due to differences technologies that are available for Systems-ADME/Tox between human and rodent physiology, genetics, metabo- (Fig. 2) along with some illustration of their applications to lism and signaling pathways. For example, the mechanism drug metabolism and drug transport which are key com- of toxicity for pyrazinamide has been reconstructed ponents of the ADME/Tox process. The ultimate aim of (Bugrim, Nikolskaya, & Nikolsky, 2004) to illustrate that this discussion is to provide awareness of an integrated the accumulation of uric acid occurs in human, but not in approach rather than a technology silo mentality, represent- mice, and this results in toxicity in the former. The relatively ing the latest proposed research model in the field (Fig. 2). poor understanding of such species differences may be reflected in the relatively large number of late stage molecules that have undergone in vivo toxicity assessment 2. Data available yet have been later withdrawn due to adverse events in humans. A recently published book provides an excellent over- In recent years the appearance of systems biology which view of toxicogenomics and the reader is referred to this to uses the relationships of all elements of biology rather than gain more insight into the applications and limitations approaching them separately has been evident and will (Hamadeh & Afshari, 2004). The growing number of likely reunite biological fields (Harrison, 2004; Hood & toxicogenomic datasets derived from in vivo studies with Galas, 2003). These systems approaches are the latest rat (Table 1) and mouse (Table 2) as well as in vitro cell incarnation of the importance of the Fparts vs wholes_ derived data (Table 3) highlights the different strains, debate (Ekins & McGowan, 2001) and interpreting ADME/ microarray types and compounds that are routinely assessed. Tox in this context may improve our understanding and Also there are numerous instances of multiple groups testing ultimate predictions (Bugrim et al., 2004; Ekins, Boulanger, the same compound at similar or different doses, e.g. well Swaan, & Hupcey, 2002a; Kitano, 2002a; Werner, 2003). known hepatotoxicants or nephrotoxicants such