NMR-based metabonomic toxicity classification: hierarchical cluster analysis and k-nearest-neighbour approaches

https://doi.org/10.1016/S0003-2670(03)00060-6Get rights and content

Abstract

The COnsortium for MEtabonomic Toxicology (COMET) project is constructing databases and metabolic models of drug toxicity using ca. 100,000 1H NMR spectra of biofluids from animals treated with model toxins. Mathematical models characterising the effects of toxins on endogenous metabolite profiles will enable rapid toxicological screening of potential drug candidates and discovery of novel mechanisms and biomarkers of specific types of toxicity.

The metabolic effects and toxicity of 19 model compounds administered to rats in separate studies at toxic (high) and sub-toxic (low) doses were investigated. Urine samples were collected from control and dosed rats at 10 time points over 8 days and were subsequently analysed by 600 MHz 1H NMR spectroscopy. In order to classify toxicity and to reveal similarities in the response of animals to different toxins, principal component analysis (PCA), hierarchical cluster analysis (HCA) and k-nearest-neighbour (kNN) classification were applied to the data from the high-dose studies to reveal dose and time-related effects. Both PCA and HCA provided valuable overviews of the data, highlighting characteristic metabolic perturbations in the urine spectra between the four groups: controls (C), liver (L) toxins, kidney (K) toxins and other (O) treatments, and revealed further differences between subgroups of liver toxins. kNN analysis of the multivariate data using both leave-one-out (LOO) cross-validation and training and test-set (50:50) classification successfully predicted all the different toxin classes. The four treatment groups (control, liver, kidney and other) were predicted with 86, 85, 91 and 88% success rate (training/test). In a study-by-study comparison, 81% of the samples were predicted into the correct toxin study (training/test). This work illustrates the high power and reliability of metabonomic data analysis using 1H NMR spectroscopy together with chemometric techniques for the exploration and prediction of toxic effects in the rat.

Introduction

In the pharmaceutical industry the early identification of toxicological side-effects of potential new drug candidates is of fundamental value in the minimisation of compound “attrition” [1]. While specific assays have been developed to detect abnormal changes in biofluids of test animals, histopathological tissue examination remains the ‘gold-standard’ in toxicological assessment. Since investigating toxicity by traditional techniques is time consuming and costly, different approaches to the reliable and effective prediction of toxicity have been explored. Currently, there are several advanced technologies using multiparametric biochemical information derived from different levels of biomolecular organisation. Genomic and proteomic information describing transcriptional effects and protein synthesis can be extracted [2], [3]. However, these approaches are also slow and expensive. Alternatively, the metabonomic approach analyses the entire pool of endogenous metabolites in a biofluid or tissue, and is a powerful technique for detecting and describing biological endpoint-effects. It has already proven to be effective in the description of genetic strain differences, disease states, nutritional effects and toxic influences [1], [4], [5], [6], [7], [8], [9]. In this study, we report the initial evaluation of metabonomic data yielded in a large-scale toxicological project within the COnsortium for MEtabonomic Toxicology (COMET). The COMET1 project is currently constructing databases and metabolic models of drug toxicity using ca. 100,000 1H NMR spectra of biofluids from rats and mice treated with approximately 150 model compounds. Multivariate statistical models characterising the effects of toxins on endogenous metabolite profiles have been generated and will be used for the development of computer-based expert systems. These will enable rapid pre-clinical toxicological screening of potential drug candidates.

As a prototype for such large and comprehensive databases we have used a subset of the data to study dose- and time-related effects, to develop and evaluate classification methods and explore characteristic features of toxicity (the ‘biomarker profile’). This facilitates the prediction and interpretation of large-scale data sets and will help to hone the development of new analytical tools. We describe a first attempt to explore and predict the effect of different toxins, surgical procedures and other external influences in 19 independent studies using 1H NMR-metabonomic analysis of rat urine samples. Urine samples were collected from dosed and control rats at 10 time points over a period of 8 days and were subsequently analysed by 600 MHz 1H NMR spectroscopy. Principal component analysis (PCA) and hierarchical cluster analysis (HCA) were applied to the data from the studies to reveal dose and time-related effects in the response of animals to different toxins. Finally, k-nearest-neighbour (kNN) classification [10] was used to predict how well individual samples were classified into their respective class and time point. This work gives an insight into the scope of this project and illustrates the diagnostic power and reliability of metabonomic data analysis using 1H NMR spectroscopy together with chemometric techniques for the exploration and prediction of toxic effects.

Section snippets

Sample collection

In established in-house or external animal research facilities of the involved pharmaceutical companies, male 8–10 week old Sprague-Dawley rats were housed in metabolism cages and urine samples collected daily. Free access to food and water was permitted throughout the studies (except in the case of food and water restriction studies). The animals were randomly assigned to three dose groups: control (vehicle only, n=10), high-dose (dose causing a clear toxic effect, n=10) and low-dose

Overall analysis

PCA of the averaged spectra at each time point gave a clear overview of the treatment- and time-related metabonomic effects as observed by NMR spectroscopy of the urine (Fig. 2). Separation of these data points can be attributed to alterations in NMR signals of endogenous urinary metabolites as seen in the loadings plot. Urines from streptozotocin-treated animals revealed a urinary metabolic profile distinct from all other treatment groups, mainly due to elevated glucose signals (δ = 3.4–3.88).

Conclusions

As part of a larger effort to construct a comprehensive database of urinary metabonomic data from toxicological investigations, these preliminary results have underlined the potential of our approach for the prediction of drug toxicity. Data were employed from animal studies showing metabonomic effects due to toxin administration, surgical procedures (partial hepatectomy, unilateral nephrectomy) and other treatments (food restriction, water deprivation). It was shown that chemometric methods

Acknowledgements

OB, MEB, TMDE and HCK thank the members of the Consortium for Metabonomic Toxicology (Bristol-Myers Squibb, Eli Lilly, Hoffmann-La Roche, Novo Nordisk, Pfizer and Pharmacia) for financial support.

References (17)

There are more references available in the full text version of this article.

Cited by (134)

  • Metabolomics approach in pharmacognosy

    2023, Pharmacognosy: Fundamentals, Applications, and Strategies, Second Edition
View all citing articles on Scopus
View full text