Systems Toxicology—a merger between toxicology and “omics”—might predict toxicity earlier in drug development.
In the wake of failures of potential blockbuster drugs late in clinical trials and post-approval, a number of initiatives have grown around the nascent field of Systems Toxicology to help pharmaceutical researchers better understand and detect potential toxicities as early as possible in drug development.
Systems Toxicology is perhaps best defined as the combination of traditional toxicological methods with novel strategies and tools for integrating data from multiple sources including high throughput transcriptomics, proteomics, and metabonomics. This approach is characterized by leveraging known conventional biomarkers and warning signs with novel informatics technologies and methods, and by the creation of innovative “cross-omics” databases.
Current advances in Systems Toxicology have begun to yield fascinating and concrete results in developing successful strategies for mechanistic and predictive toxicology. One approach to find new predictive biomarkers is to re-investigate failed drugs using a combination of cross-omics and classical toxicological endpoint analyses to detect toxicants earlier in the process.
In predictive studies, the goal is to seek markers to evaluate the likelihood of whether a certain compound will be toxic to humans. Thus, the obvious primary objective is to identify the toxicity of a compound with high accuracy and specificity. To achieve this, the first goal is to build a reference “library” to identify biomarkers based on studies on a set of compounds with a known toxicological profile.
The second objective is to gain a better understanding of the specific mode of action (MOA) of a toxicant, rather than to provide a simple binary toxic/non-toxic assignment. This approach can also help to predict what kind of toxicity a compound might induce, and whether it is relevant to humans in case the reference library is based on an animal model.
Given the immense number of possible signals, effective toxicity prediction and understanding of MOA depend on the use of specific biomarkers for initial identification and subsequent development of a robust diagnostic assay.
The initial steps are identical to conventional toxicology studies:
• Animals are treated with a compound under controlled conditions. The study schedule may involve one or several applications of the compound under examination as well as various biological sampling steps. Eventually, treated and control subjects undergo necropsy including full pathological assessment.
• During the –omics (molecular profiling) phase, predefined tissue samples (e.g., liver, kidney) and blood from the animals are subjected to any combination of -omics investigations, involving transcriptomics, proteomics, and metabonomics.
• In the subsequent bioinformatics phase, the resulting data is processed and analyzed either separately for the different data types or in an integrated fashion. This gives rise to models describing the MOA of a particular compound or sets of markers that can be used to predict the toxic potential of further compounds.
• The results of the analysis are then slated for validation during follow-up studies, for potential use in routine operations.
The basic outline for toxicogenomics approaches to Systems Toxicology solutions is simple, even elegant. However, it is clear that the challenge to realizing the potential lies in the integration and unification of a multiplicity of data resources to generate useful knowledge. In addition to the problem of sifting vast masses of data, researchers must overcome the complexities of multiple and incompatible sources: data types and formats, databases, and tools for analysis.
Overcoming these hurdles requires new technologies that enable workflows for efficient capture, storage, integrated analysis and comparison of conventional tox and various -omics data, not merely for one research group, but across a diverse range of scientific groups, bench scientists, and decision-makers at pharmaceutical and biotechnology companies. In the course of this requirements analysis, four necessary supporting technologies were identified:
• Integration across varied data types and formats
• Integration with major high throughput technology platforms
• Novel standardization and automation methods for processing and analysis
• A full-spectrum workflow, from experimental design to completed study reports
InnoMed PredTox consists of multiple global biopharmas, academic centers and bio-IT companies, including: |
Pharma/Biotech Companies |
Bayer Healthcare |
Bayer Schering Pharma |
Boehringer Ingelheim |
Firmenprofil |
Johnson & Johnson |
Lilly |
Merck KGaA |
Merck Serono/RBM |
Novartis |
Novo Nordisk |
Nycomed |
Organon |
Roche |
Sanofi-Aventis |
Servier |
University Research |
University College Dublin (Ireland) |
University of Hacettepe (Turkey) |
University of Wuerzburg (Germany) |
Technology Providers |
Bio-Rad Laboratories Inc. |
Genedata |
Over the past three years, international collaborations in Systems Toxicology have made great strides in meeting both the overall toxicological screening objectives and the proposed technology and workflow goals. The InnoMed PredTox Consortium (Figure 2), coordinated by the European Federation of Pharmaceutical Industries and Associations (EFPIA), has been a leader in this area.
InnoMed PredTox examined compounds that failed during non-clinical development. The overarching objectives were to engineer innovative methods for catching these failures, retroactively, and to develop technologies to be eventually applied to current pipelines. The chief rationale, of course, is to help prevent future late failures and their associated economic and lost opportunity costs. The Consortium characterized 14 proprietary compounds contributed by the individual member companies. Those compounds were dropped from development after they had failed conventional toxicology tests. In addition, two reference compounds with known toxic profiles, Troglitazone and Gentamicin, were chosen to complement the project library.
The experimental design of InnoMed PredTox is unique with respect to the number of assay technologies that are included to assess toxicity of the study compounds. In response, Genedata shaped both the database and the data analysis system to bring together all the resulting data types while recognizing and modeling their peculiarities.
The key hurdles faced by InnoMed PredTox included:
• Standardization of protocols
• Animal treatment
• Dosing regimen
• Sample requirements and preparation
• Classical endpoint-controlled vocabularies
• Technology-specific assays
• Comparability and reproducibility of assays
• Between sites
• Across technology platforms
• Across molecular levels (e.g., transcripts and protein expression, metabolite abundances)
• Consensus on the data analysis strategy and the meaning of the markers
Within the first two years of InnoMed PredTox, consortium-wide procedural and experimental standards were developed, animal toxicity studies performed, and the generated data (transcriptomics, proteomics, metabonomics, and conventional toxicology) subsequently processed and stored in PredTox DB. This database is the consortium-wide central infrastructure created to capture all collected data and store it in a relational fashion.
Within the first months of 2008, collection of experimental data and data analysis for the individual compound studies were completed.
The experimental study design for the Systems Toxicology effort featured:
• in vivo animal studies in rats
• 16 compounds with known hepato- and/or nephrotoxicity
• Two reference compounds (one for each toxicity type)
• 14 proprietary compounds that dropped out of development
• Study design
• Two dose levels and (time-matched) vehicle control groups
• Three evaluation time points (24 hour, three days, and 14 days)
• Investigations
• Primary targets: liver, kidney, blood, urine
• “Classical” endpoints in toxicology studies (e.g., clinical chemistry, histopathology)
• Omics technologies
• Transcriptomics
(Affymetrix IVT arrays)
• Proteomics (2D-DIGE, 2D-PAGE and SELDI)
• Metabonomics (LC/MS and NMR)
In accordance with the study requirements, PredTox DB captures and stores relevant documents, sample annotation details, measured data, and platform-specific information. Being built around the “biological subject” (e.g., animals, samples, or aliquots), the database closely models the experimental design of systematic in vivo toxicology studies.
The data processing, standardization, and analysis workflow was created by technology-specific quality assessment, pre-processing and normalization of the omics data, followed by standardized statistical analysis steps for each study, and by specific cross-compound comparisons.
The overall technical quality of the microarray data was very good: Less than 2% of the greater than 2,000 microarrays showed likely experimental artifacts that required re-hybridization. For the large proteomics and metabonomics datasets generated with LC-MS (liquid chromatography-coupled mass spectrometry), noise subtraction, retention time alignment, peak extraction, and peak shaping was performed. This led to high-quality data sets containing several thousands of peaks per study. Similarly, clean peak lists were obtained from the proteomics datasets obtained with surface-enhanced laser desorption ionization technology (SELDI).
In order to identify discriminative transcripts, metabolites, proteins, and conventional toxicology items, several statistical methods were applied to the datasets. These included uni- and multivariate analyses like Principal Components Analysis (PCA), N-way Analysis of Variance (ANOVA), and two-group comparisons like t-Tests and the computation of fold changes. The results were used to generate and test hypotheses on the molecular mechanism that cause the toxic effects of each compound.
Currently, and in the years to come, traditional and emerging approaches to toxicology evaluation will continue. In order to define meaningful categories for prediction and to enable biological understanding of molecular findings, it is important for researchers to integrate as much of the available information on conventional readouts (e.g., blood chemistry, histopathology) and jointly analyze the data. Subsequently, the quality of prediction based on Systems Toxicology efforts will increase with the size and the level of associated information of the reference libraries, and with the emerging novel informatics and workflows designed around them.
About the Author
Jochen König has more than 10 years experience in molecular biology research and managing scientific and technical collaborations. He focuses on application of “-omics” in toxicology and diagnostics, and is coordinating Genedata’s participation in international research consortia.
This article was published in Drug Discovery & Development magazine: Vol. 11, No. 8, August, 2008, pp. 28-30.
Filed Under: Drug Discovery