Classification of toxin-induced changes in 1H NMR spectra of urine using an artificial neural network

The application of artificial neural networks in metabolomics: a historical perspective

BACKGROUND

Metabolomics data, with its complex covariance structure, is typically modelled by projection-based machine learning (ML) methods such as partial least squares (PLS) regression, which project data into a latent structure. Biological data are often non-linear, so it is reasonable to hypothesize that metabolomics data may also have a non-linear latent structure, which in turn would be best modelled using non-linear equations. A non-linear ML method with a similar projection equation structure to PLS is artificial neural networks (ANNs). While ANNs were first applied to metabolic profiling data in the 1990s, the lack of community acceptance combined with limitations in computational capacity and the lack of volume of data for robust non-linear model optimisation inhibited their widespread use. Due to recent advances in computational power, modelling improvements, community acceptance, and the more demanding needs for data science, ANNs have made a recent resurgence in interest across research communities, including a small yet growing usage in metabolomics. As metabolomics experiments become more complex and start to be integrated with other omics data, there is potential for ANNs to become a viable alternative to linear projection methods.

AIM OF REVIEW

We aim to first describe ANNs and their structural equivalence to linear projection-based methods, including PLS regression. We then review the historical, current, and future uses of ANNs in the field of metabolomics.

KEY SCIENTIFIC CONCEPT OF REVIEW

Is metabolomics ready for the return of artificial neural networks?

Metabolic changes associated with papillary thyroid carcinoma: A nuclear magnetic resonance-based metabolomics study

Papillary thyroid carcinoma (PTC) is the most common thyroid cancer. Nuclear magnetic resonance (NMR)‑based metabolomic technique is the gold standard in metabolite structural elucidation, and can provide different coverage of information compared with other metabolomic techniques. Here, we firstly conducted NMR based metabolomics study regarding detailed metabolic changes especially metabolic pathway changes related to PTC pathogenesis. 1H NMR-based metabolomic technique was adopted in conju-nction with multivariate analysis to analyze matched tumor and normal thyroid tissues obtained from 16 patients. The results were further annotated with Kyoto Encyclopedia of Genes and Genomes (KEGG), and Human Metabolome Database, and then were analyzed using modules of pathway analysis and enrichment analysis of MetaboAnalyst 3.0. Based on the analytical techniques, we established the models of principal component analysis (PCA), partial least squares-discriminant analysis (PLS-DA), and orthogonal partial least-squares discriminant analysis (OPLS‑DA) which could discriminate PTC from normal thyroid tissue, and found 15 robust differentiated metabolites from two OPLS-DA models. We identified 8 KEGG pathways and 3 pathways of small molecular pathway database which were significantly related to PTC by using pathway analysis and enrichment analysis, respectively, through which we identified metabolisms related to PTC including branched chain amino acid metabolism (leucine and valine), other amino acid metabolism (glycine and taurine), glycolysis (lactate), tricarboxylic acid cycle (citrate), choline metabolism (choline, ethanolamine and glycerolphosphocholine) and lipid metabolism (very-low‑density lipoprotein and low-density lipoprotein). In conclusion, the PTC was characterized with increased glycolysis and inhibited tricarboxylic acid cycle, increased oncogenic amino acids as well as abnormal choline and lipid metabolism. The findings in this study provide new insights into detailed metabolic changes of PTC, and hold great potential in the treatment of PTC.

Machine learning algorithms for mode-of-action classification in toxicity assessment

Background

Real Time Cell Analysis (RTCA) technology is used to monitor cellular changes continuously over the entire exposure period. Combining with different testing concentrations, the profiles have potential in probing the mode of action (MOA) of the testing substances.

Results

In this paper, we present machine learning approaches for MOA assessment. Computational tools based on artificial neural network (ANN) and support vector machine (SVM) are developed to analyze the time-concentration response curves (TCRCs) of human cell lines responding to tested chemicals. The techniques are capable of learning data from given TCRCs with known MOA information and then making MOA classification for the unknown toxicity. A novel data processing step based on wavelet transform is introduced to extract important features from the original TCRC data. From the dose response curves, time interval leading to higher classification success rate can be selected as input to enhance the performance of the machine learning algorithm. This is particularly helpful when handling cases with limited and imbalanced data. The validation of the proposed method is demonstrated by the supervised learning algorithm applied to the exposure data of HepG2 cell line to 63 chemicals with 11 concentrations in each test case. Classification success rate in the range of 85 to 95 % are obtained using SVM for MOA classification with two clusters to cases up to four clusters.

Conclusions

Wavelet transform is capable of capturing important features of TCRCs for MOA classification. The proposed SVM scheme incorporated with wavelet transform has a great potential for large scale MOA classification and high-through output chemical screening.

Electronic supplementary material

The online version of this article (doi:10.1186/s13040-016-0098-0) contains supplementary material, which is available to authorized users.

The role of metabolic biomarkers in drug toxicity studies

ABSTRACT Metabolic profiling is a technique that can potentially provide more sensitive and specific biomarkers of toxicity than the current clinical measures benefiting preclinical and clinical drug studies. Both nuclear magnetic resonance (NMR) and mass spectrometry (MS) platforms have been used for metabolic profiling studies of drug toxicity. Not only can both techniques provide novel biomarker(s) of toxicity but the combination of both techniques gives a broader range of metabolites evaluated. Changes in metabolic patterns can provide insight into mechanism(s) of toxicity and help to eliminate a potentially toxic new chemical entity earlier in the developmental process. Metabolic profiling offers numerous advantages in toxicological research and screening as sample collection and preparation are relatively simple. Further, sample throughput, reproducibility, and accuracy are high. The area of drug toxicity of therapeutic compounds has already been impacted by metabolic profiling studies and will continue to be impacted as new, more specific biomarker(s) are found. In order for a biomarker or pattern of biomarkers to be accepted, it must be shown that they originate from the target tissue of interest. Metabolic profiling studies are amenable to any biofluid or tissue sample making it possible to link the changes noted in urine for instance as originating from renal injury. Additionally, the ease of sample collection makes it possible to follow a single animal or subject over time in order to determine whether and when the toxicity resolves itself. This review focuses on the advantages of metabolic profiling for drug toxicity studies.

Identification of serum biomarkers for lung cancer using magnetic bead-based SELDI-TOF-MS

AIM

To identify novel serum biomarkers for lung cancer diagnosis using magnetic bead-based surface-enhanced laser desorption/ionization time-of-flight mass spectrum (SELDI-TOF-MS).

METHODS

The protein fractions of 121 serum specimens from 30 lung cancer patients, 30 pulmonary tuberculosis patients and 33 healthy controls were enriched using WCX magnetic beads and subjected to SELDI-TOF-MS. The spectra were analyzed using Bio-marker Wizard version 3.1.0 and Biomarker Patterns Software version 5.0. A diagnostic model was constructed with the marker proteins using a linear discrimination analysis method. The validity of this model was tested in a blind test set consisted of 8 randomly selected lung cancer patients, 10 pulmonary tuberculosis patients and 10 healthy volunteers.

RESULTS

Seventeen m/z peaks were identified, which were significantly different between the lung cancer group and the control (tuberculosis and healthy control) groups. Among these peaks, the 6445, 9725, 11705, and 15126 m/z peaks were selected by the Biomarker Pattern Software to construct a diagnostic model for lung cancer. This four-peak model established in the training set could discriminate lung cancer patients from non-cancer patients with a sensitivity of 93.3% (28/30) and a specificity of 90.5% (57/63). The diagnostic model showed a high sensitivity (75.0%) and a high specificity (95%) in the blind test validation. Database searching and literature mining indicated that the featured 4 peaks represented chaperonin (M9725), hemoglobin subunit beta (M15335), serum amyloid A (M11548), and an unknown protein.

CONCLUSION

A lung cancer diagnostic model based on bead-based SELDI-TOF-MS has been established for the early diagnosis or differential diagnosis of lung cancers.

Identification of serum biomarkers of hepatocarcinoma through liquid chromatography/mass spectrometry-based metabonomic method

Late diagnosis of hepatocarcinoma (HCC) is one of the most primary factors for the poor survival of patients. Thereby, identification of sensitive and specific biomarkers for HCC early diagnosis is of great importance in biological medicine to date. In the present study, serum metabolites of the HCC patients and healthy controls were investigated using the improved liquid chromatography-mass spectrometry (LC/MS). A wavelet-based method was utilized to find and align peaks of LC-MS. The characteristic peaks were selected by performing a two-sample t test statistics (p value <0.05). Clustering analysis based on principal component analysis showed a clear separation between HCC patients and healthy individuals. The serum metabolite, namely 1-methyladenosine, was identified as the characteristic metabolite for HCC. Moreover, receiver-operator curves were calculated with 1-methyladenosine and/or alpha fetal protein (AFP). The higher area under curve value was achieved in 1-methyladenosine group than AFP group (0.802 vs. 0.592), and the diagnostic model combining 1-methyladenosine with AFP exhibited significant improved sensitivity, which could identify those patients who missed the diagnosis of HCC by determining serum AFP alone. Overall, these results suggested that LC/MS-based metabonomic study is a potent and promising strategy for identifying novel biomarkers of HCC.

GC/MS-based metabolomic approach to validate the role of urinary sarcosine and target biomarkers for human prostate cancer by microwave-assisted derivatization

A recent study showed that sarcosine may be potentially useful for the diagnosis and prognosis of prostate cancer (PCa). The aim of this study was to validate diagnostic value of sarcosine for PCa, to evaluate urine metabolomic profiles in patients with PCa in comparison of non-cancerous control, and to further explore the other potential metabolic biomarkers for PCa. Isotope dilution gas chromatography/mass spectrometry (ID GC/MS) metabolomic approach was applied to evaluate sarcosine using [methyl-D(3)]-sarcosine as an internal standard. Microwave-assisted derivatization (MAD) together with GC/MS was utilized to obtain the urinary metabolomic information in 20 PCa patients compared with eight patients with benign prostate hypertrophy and 20 healthy men. Acquired metabolomic data were analyzed using a two-sample t test. Diagnostic models for PCa were constructed using principal component analysis and were assessed with receiver-operating characteristic curves. Results showed that the urinary sarcosine level has no statistical difference between the PCa group and the control group. In addition, nine metabolomic markers between the PCa group and the healthy male group were selected, which constructed a diagnostic model with a high area under the curve value of 0.9425. We conclude that although urinary sarcosine value has limited potential in the diagnostic algorithm of PCa, urinary metabolomic panel based on GC/MS assay following MAD may potentially become a diagnostic tool for PCa.

Metabolomics for early detection of drug-induced kidney injury: review of the current status

The identification of biomarkers of drug-induced kidney injury is an area of intensive focus in drug development. Traditional markers of renal function, including blood urea nitrogen and serum creatinine, are not region-specific and only increase significantly after substantial kidney injury. Therefore, more sensitive markers of kidney injury are needed. The ideal biomarkers will identify nephrotoxicity early in the drug-discovery process, resulting in decreased development costs and safer drugs. Metabolomics, the study of the small biochemicals present in a biological sample, has become a promising player in the nephrotoxicity arena. In this review, we describe the current status of the identification of metabolic biomarkers for drug-induced kidney toxicity screening. Many of these markers have been confirmed across multiple studies and can detect nephrotoxicity earlier than the traditional clinical chemistry and histopathology methods. Upon further validation, such markers will offer clear benefits for the pharmaceutical industry and regulatory agencies.

Metabolomic investigation of gastric cancer tissue using gas chromatography/mass spectrometry

Gastric cancer screening or diagnosis is mainly based on endoscopy and biopsy. The aim of this study was to identify the difference of metabolomic profile between normal and malignant gastric tissue, and to further explore tumor biomarkers. Chemical derivatization together with gas chromatography/mass spectrometry (GC/MS) was utilized to obtain the metabolomic information of the malignant and non-malignant tissues of gastric mucosae in 18 gastric cancer patients. Acquired metabolomic data was analyzed using the Wilcoxon rank sum test to find the tissue metabolic biomarkers for gastric cancer. A diagnostic model for gastric cancer was constructed using principal component analysis (PCA), and was assessed with receiver-operating characteristic (ROC) curves. Results showed that 18 metabolites were detected differently between the malignant tissues and the adjacent non-malignant tissues of gastric mucosa. Five metabolites were also detected differently between the non-invasive tumors and the invasive tumors. The diagnostic model could discriminate tumors from normal mucosae with an area under the curve (AUC) value of 0.9629, and another diagnostic model constructed for clinical staging was assessed with an AUC value of 0.969. We conclude that the metabolomic profile of malignant gastric tissue was different from normal, and that the selected tissue metabolites could probably be applied for clinical diagnosis or staging for gastric cancer.

Mixture analysis by NMR as applied to fruit juice quality control

Nuclear magnetic resonance (NMR) spectroscopy is rapidly gaining importance in mixture analysis, originally driven by the pharmaceutical and nowadays also by clinical applications within metabonomics. Quality control of food-related material has very similar requirements, as it also deals with mixtures, and many of the compounds found in body fluids are analyzed as well. NMR allows analysis in two ways within one experiment: namely, targeted and untargeted. Targeted stands for the safe identification and consequent quantification of individual compounds, whereas untargeted means the detection of all deviations visible by NMR using statistical analysis based on normality models. Very important is the stability and reproducibility of the NMR instrumentation used, and this means inherent minimized system internal variance. NMR is especially suited for such requirements, as it allows detection of the smallest concentration changes of many metabolites simultaneously. High-throughput flow-injection NMR as the basis for fruit juice screening allows low cost per sample and delivers substantially more relevant information than any other method and is probably the only method to produce such results.

Metabolomic study for diagnostic model of oesophageal cancer using gas chromatography/mass spectrometry

The prognosis for oesophageal cancer is poor. Attempts have been made for the identification of biomarkers for early diagnosis. Metabolomic panel has been evaluated as potential candidate biomarkers. With gas chromatography/mass spectrometry (GC/MS) as a sensitive modality for metabolomics, various tissue metabolites can be detected and identified. We hypothesized that tissue metabolomic biomarkers may be identifiable and diagnostically useful for oesophageal cancer. We present a metabolomic method of chemical derivatization followed by GC/MS to analyze the metabolic difference in biopsied specimens between oesophageal cancer and corresponding normal mucosae obtained from 20 oesophageal cancer patients. The GC/MS data was analyzed using a two sample t-test to explore the potential metabolic biomarkers for oesophageal cancer. A diagnostic model was constructed to discriminate normal from malignant samples, using principal component analysis (PCA) and receiver-operating characteristic (ROC) curves. t-Test showed a total of 20 marker metabolites detected were found to be different with statistical significance (P<0.05). The multivariate logistic analysis yielded a complete distinction between the two groups. The diagnostic model could discriminate tumors from normal mucosae with an area under the curve (AUC) value of 1. Our findings suggest that this assay may potentially provide a new metabolomic biomarker for oesophageal cancer.

Metabolomic profiling of human urine in hepatocellular carcinoma patients using gas chromatography/mass spectrometry

With the technique of metabolomics, gas chromatography/mass spectrometry (GC/MS), urine or serum metabolites can be assayed to explore disease biomarkers. In this work, we present a metabolomic method to investigate the urinary metabolic difference between hepatocellular carcinoma (HCC, n - 20) male patients and normal male subjects (n - 20). The urinary endogenous metabolome was assayed using chemical derivatization followed by GC/MS. After GC/MS analysis, 103 metabolites were detected, of which 66 were annotated as known compounds. By a two sample t-test statistics with p < 0.05, 18 metabolites were shown to be significantly different between the HCC and control groups. A diagnostic model was constructed with a combination of 18 marker metabolites or together with alphafetoprotein, using principal component analysis and receiver-operator characteristic curves. The multivariate statistics of the diagnostic model yielded a separation between the two groups with an area under the curve value of 0.9275. This non-invasive technique of identifying HCC biomarkers from urine may have clinical utility.

Global metabolic profiling and its role in systems biology to advance personalized medicine in the 21st century

Systems biology attempts to elucidate the complex interaction between genes, proteins and metabolites to provide a mechanistic understanding of cellular function and how this function is affected by disease processes, drug toxicity or drug efficacy effects. Global metabolic profiling is an important component of systems biology that can be applied in both preclinical and clinical settings for drug discovery and development, and to study disease mechanisms. The metabolic profile encodes the phenotype, which is composed of the genotype and environmental factors. The phenotypic profile can be used to make decisions about the best course of treatment for an individual patient. Understanding the combined effects of genetics and environment through a systems biology framework will enable the advancement of personalized medicine.

Genetic algorithms for simultaneous variable and sample selection in metabonomics

MOTIVATION

Metabolic profiles derived from high resolution (1)H-NMR data are complex, therefore statistical and machine learning approaches are vital for extracting useful information and biological insights. Focused modelling on targeted subsets of metabolites and samples can improve the predictive ability of models, and techniques such as genetic algorithms (GAs) have a proven utility in feature selection problems. The Consortium for Metabonomic Toxicology (COMET) obtained temporal NMR spectra of urine from rats treated with model toxins and stressors. Here, we develop a GA approach which simultaneously selects sets of samples and spectral regions from the COMET database to build robust, predictive classifiers of liver and kidney toxicity.

RESULTS

The results indicate that using simultaneous sample and variable selection improved performance by over 9% compared with either method alone. Simultaneous selection also halved computation time. Successful classifiers repeatedly selected particular variables indicating that this approach can aid defining biomarkers of toxicity. Novel visualizations of the results from multiple computations were developed to aid the interpretability of which samples and variables were frequently selected. This method provides an efficient way to determine the most discriminatory variables and samples for any post-genomic dataset.

AVAILABILITY

GA code available from http://www1.imperial.ac.uk/medicine/people/r.cavill/

Analysis of urinary metabolites for metabolomic study by pressurized CEC

A new approach for the metabolomic study of urinary samples using pressurized CEC (pCEC) with gradient elution is proposed as an alternative chromatographic separation tool with higher degree of resolution, selectivity, sensitivity, and efficiency. The pCEC separation of urinary samples was performed on a RP column packed with C(18), 5 microm particles with an ACN/water mobile phase containing TFA. The effects of the acid modifiers, applied voltage, mobile phase, and detection wavelength were systematically evaluated using eight spiked standards, as well as urine samples. A typical analytical trial of urine samples from Sprague Dawley (S.D.) rats exposed to high-energy diet was carried out following sample pretreatment. Significant differences in urinary metabolic profiles were observed between the high energy diet-induced obesity rats and the healthy control rats at the 6th wk postdose. Multivariate statistical analysis revealed the differential metabolites in response to the diet, which were partially validated with the putative standards. This work suggests that such a pCEC-based separation and analysis method may provide a new and cost-effective platform for metabolomic study uniquely positioned between the conventional chromatographic tools such as HPLC, and hyphenated analytical techniques such as LC-MS.

1H NMR metabonomics of plasma lipoprotein subclasses: elucidation of metabolic clustering by self-organising maps

(1)H NMR spectra of plasma are known to provide specific information on lipoprotein subclasses in the form of complex overlapping resonances. A combination of (1)H NMR and self-organising map (SOM) analysis was applied to investigate if automated characterisation of subclass-related metabolic interactions can be achieved. To reliably assess the intrinsic capability of (1)H NMR for resolving lipoprotein subclass profiles, sum spectra representing the pure lipoprotein subclass part of actual plasma were simulated with the aid of experimentally derived model signals for 11 distinct lipoprotein subclasses. Two biochemically characteristic categories of spectra, representing normolipidaemic and metabolic syndrome status, were generated with corresponding lipoprotein subclass profiles. A set of spectra representing a metabolic pathway between the two categories was also generated. The SOM analysis, based solely on the aliphatic resonances of these simulated spectra, clearly revealed the lipoprotein subclass profiles and their changes. Comparable SOM analysis in a group of 69 experimental (1)H NMR spectra of serum samples, which according to biochemical analyses represented a wide range of lipoprotein lipid concentrations, corroborated the findings based on the simulated data. Interestingly, the choline-N(CH(3))(3) region seems to provide more resolved clustering of lipoprotein subclasses in the SOM analyses than the methyl-CH(3) region commonly used for subclass quantification. The results illustrate the inherent suitability of (1)H NMR metabonomics for automated studies of lipoprotein subclass-related metabolism and demonstrate the power of SOM analysis in an extensive and representative case of (1)H NMR metabonomics.

LC-MS-based metabonomics analysis

Metabonomics aims at the comprehensive and quantitative analysis of wide arrays of metabolites in biological samples. It has shown particular promise in the areas of toxicology and drug development, functional genomics, systems biology, and clinical diagnosis. Comprehensive metabonomics investigations are primarily a challenge for analytical chemistry. High-performance liquid chromatography-mass spectrometry (HPLC-MS) is an established technology in drug metabolite analysis and is now expanding into endogenous metabolite research. Its main advantages include wide dynamic range, reproducible quantitative analysis, and the ability to analyze biofluids with extreme molecular complexity. The aims of developing HPLC-MS for metabonomics range from understanding basic biochemistry to biomarker discovery and the structural characterization of physiologically important metabolites. In this review, the strategy and application of HPLC-MS-based metabonomics are reviewed.

Prediction and classification of drug toxicity using probabilistic modeling of temporal metabolic data: the consortium on metabonomic toxicology screening approach

Detection and classification of in vivo drug toxicity is an expensive and time-consuming process. Metabolic profiling is becoming a key enabling tool in this area as it provides a unique perspective on the characterization and mechanisms of response to toxic insult. As part of the Consortium on Metabonomic Toxicology (COMET) project, a substantial metabolic and pathological database was constructed. We chose a set of 80 treatments to build a modeling system for toxicity prediction using NMR spectroscopy of urine samples (n=12935) from laboratory rats (n=1652). The compound structures and activities were diverse but there was an emphasis on the selection of hepato and nephrotoxins. We developed a two-stage strategy based on the assumptions that (a) adverse effects would produce metabolic profiles deviating from those of normal animals and (b) such deviations would be similar for treatments having similar physiological effects. To address the first stage, we developed a multivariate model of normal urine, using principal components analysis of specially preprocessed 1H NMR spectra. The model demonstrated a high correspondence between the occurrence of toxicity and abnormal metabolic profiles. In the second stage, we extended a density estimation method, "CLOUDS", to compute multidimensional similarities between treatments. Crucially, the technique allowed a distribution-free estimate of similarity across multiple animals and time points for each treatment and the resulting matrix of similarities showed segregation between liver toxins and other treatments. Using the similarity matrix, we were able to correctly identify the target organ of two "blind" treatments, even at sub-toxic levels. To further validate the approach, we then applied a leave-one-out approach to predict the main organ of toxicity (liver or kidney) showing significant responses using the three most similar matches in the matrix. Where predictions could be made, there was an error rate of 8%. The sensitivities to liver and kidney toxicity were 67 and 41%, respectively, whereas the corresponding specificities were 77 and 100%. In some cases, it was not possible to make predictions because of interference by drug-related metabolite signals (18%), an inconsistent histopathological or urinary response (11%), genuine class overlap (8%), or lack of similarity to any other treatment (2%). This study constitutes the largest validation to date of the metabonomic approach to preclinical toxicology assessment, confirming that the methodology offers practical utility for rapid in vivo drug toxicity screening.

Metabolic profiling analysis of a D-galactosamine/lipopolysaccharide-induced mouse model of fulminant hepatic failure

The purpose of this study was to characterize the changes in metabolic intermediates and to investigate the metabolic profile of a mouse model of fulminant hepatic failure (FHF), induced by D-galactosamine/lipopolysaccharide (GalN/LPS). Plasma metabolite levels were detected using gas chromatography/time-of-flight mass spectrometry, and the acquired data were transferred into Simca-P and processed using principal components analysis (PCA). In total, 45 metabolites were identified from the 267 distinct compounds found in the study. Whereas significant differences were noted in the plasma levels of the control and FHF groups, no differences in gluconeogenesis or glycolysis were noted following GalN/LPS treatment. Our data also suggest that the production of ketone bodies, and the tricarboxylic acid and urea cycles, was inhibited. PCA data suggest that 5-hydroxyindoleacetic acid, glucose, beta-hydroxybutyrate, and phosphate parameters had the highest weights on each of the principal components, and that they were the most important metabolites contributing to the separation of groups. In conclusion, this metabonomic approach can be used as a powerful tool to characterize changes in metabolic intermediates and to search for metabolic markers under certain pathophysiological conditions, such as FHF. Our data also demonstrate that a combination of 5-hydroxyindoleacetic acid, glucose, beta-hydroxybutyrate, and phosphate concentrations in the plasma is a potential marker for FHF, as well as for the early prognosis of FHF.

Metabonomics in pharmaceutical discovery and development

Metabonomics has emerged as a key technology in pharmaceutical discovery and development, evolving as the small molecule counterpart of transcriptomics and proteomics. In drug discovery laboratories, metabonomics aids in target identification, phenotyping, and the understanding of the biochemical basis of disease and toxicity. This review focuses on three areas where metabonomics is used in the industry: (1) analytical considerations, (2) chemometric and statistical concerns, and (3) biological aspects and applications.

[Advances in "omics" technologies for toxicological research]

Toxicology research can be applied to evaluate potential human health risks resulting from exposure to chemicals and other factors in the environment. The tremendous advances that have been made in high-throughput "omics" technologies (e.g., genomics, transcriptomics, proteomics and metabolomics) are providing good tools for toxicological research. Toxicogenomics is the study of changes in gene expression, protein and metabolite profiles, and combines the tools of traditional toxicology with those of genomics and bioinformatics. In particular, identification of changes in gene expression using DNA microarrays is an important method for understanding toxicological processes and obtaining an informative biomarker. Although these technologies have emerged as a powerful tool for clarifying hazard mechanisms, there are some concerns for the application of these technologies to toxicological research. This review summarizes the impact of "omics" technologies in toxicological study, followed by a brief discussion of future research.

Metabolic profiling technologies for biomarker discovery in biomedicine and drug development

The state-of-the-art of nuclear magnetic resonance spectroscopy, mass spectrometry and statistical tools for the acquisition and evaluation of complex multidimensional spectroscopic data in metabolic profiling is reviewed in this article. The continuous evolution of the sensitivity, precision and throughput has made these technologies powerful and extremely robust tools for application in systems biology, pharmaceutical and diagnostics research. Particular emphasis is also given to the collection and storage of biological samples that are subjected to metabolite profiling. Selected examples from preclinical and clinical applications are paradigmatically shown. These illustrate the power of the profiling technologies for characterizing the metabolic phenotype of healthy, diseased and treated subjects. The complexity of disease and drug treatment is asking for an adequate response by integrated and comprehensive metabolite profiling approaches that allow the discovery of new combinations of metabolic biomarkers.

Metabonomic study on the biochemical profiles of a hydrocortisone-induced animal model

This work describes the metabonomic study of a biochemical modification in vivo induced by high dose of hydrocortisone, which led to a unique pathologic condition similar to the 'kidney deficiency syndromes', an early stage of obesity and diabetes in traditional Chinese medicine. The methodology of the metabonomic approach consisted of GC/MS and multivariate statistical technique for the establishment of urine metabolic patterns of the treatment rats. In the study, 24-h urine was collected pre-dose and at days 1, 3, 7, and 10 post-dose after rats were injected with hydrocortisone at 1.5 mg/100 g. The acquired data were transferred into Matlab to be processed using principal components analysis (PCA). The results indicated that clear and consistent biochemical changes following hydrocortisone intervention under controlled conditions could be identified using chemometric analysis. The work suggests that this metabonomic approach could be used as a potentially powerful tool to investigate the biochemical changes of certain physiopathologic conditions such as metabolic syndrome, as an early diagnostic means.

A robust clustering approach for NMR spectra of natural product extracts

A robust method was developed to cluster similar NMR spectra from partially purified extracts obtained from a range of marine sponges and a plant biota. The NMR data were acquired using microtiter plate NMR (VAST) in protonated solvents. A sample data set which contained several clusters was used to optimize the protocol. The evaluation of the robustness was performed using three different clustering methods: tree clustering analysis, K-means clustering and multidimensional scaling. These methods were compared for consistency using the sample data set and the optimized methodology was applied to clustering of a set of spectra from partially purified biota extracts.

Metabonomics in Toxicology: A Review

Metabonomics and its many pseudonyms (metabolomics, metabolic profiling, etc.) have exploded onto the scientific scene in the past 2 to 3 years. Nowhere has the impact been more profound than within the toxicology community. Within this community there exists a great deal of uncertainty about whether metabonomics is something to count on or just the most recent technological flash in the pan. Much of the uncertainty is due to unfamiliarity with analytical and chemometric facets of the technology and the attendant fear of any "black-box." With those fears in mind, metabonomics technology is reviewed with particular emphasis on toxicologic applications in preclinical drug development. The jargon, logistics, and applications of the technology are covered in some detail with emphasis on recent work in the field.

Metabonomics technologies and their applications in physiological monitoring, drug safety assessment and disease diagnosis

In this review, metabonomics, a combination of data-rich analytical chemical measurements and chemometrics for profiling metabolism in complex systems, is described and its applications are reviewed. Metabonomics is typically carried out using biofluids or tissue samples. The relevance of the technique is reviewed in relation to other '-omics', and it is shown how the methods can be applied to physiological evaluation, drug safety assessment, characterization of genetically modified animal models of disease, diagnosis of human disease, and drug therapy monitoring. The different types of analytical data, mainly from nuclear magnetic resonance spectroscopy and mass spectrometry, are summarized. The outputs from a metabonomics study allow sample classification, for example according to phenotype, drug safety or disease diagnosis, and interpretation of the reasons for classification yields information on combination biomarkers of effect. Transcriptomic and metabonomic data is currently being further integrated into a holistic understanding of systems biology. An assessment of the possible future role and impact of metabonomics is presented.

Contemporary issues in toxicology the role of metabonomics in toxicology and its evaluation by the COMET project

The role that metabonomics has in the evaluation of xenobiotic toxicity studies is presented here together with a brief summary of published studies. To provide a comprehensive assessment of this approach, the Consortium for Metabonomic Toxicology (COMET) has been formed between six pharmaceutical companies and Imperial College of Science, Technology and Medicine (IC), London, UK. The objective of this group is to define methodologies and to apply metabonomic data generated using (1)H NMR spectroscopy of urine and blood serum for preclinical toxicological screening of candidate drugs. This is being achieved by generating databases of results for a wide range of model toxins which serve as the raw material for computer-based expert systems for toxicity prediction. The project progress on the generation of comprehensive metabonomic databases and multivariate statistical models for prediction of toxicity, initially for liver and kidney toxicity in the rat and mouse, is reported. Additionally, both the analytical and biological variation which might arise through the use of metabonomics has been evaluated. An evaluation of intersite NMR analytical reproducibility has revealed a high degree of robustness. Second, a detailed comparison has been made of the ability of the six companies to provide consistent urine and serum samples using a study of the toxicity of hydrazine at two doses in the male rat, this study showing a high degree of consistency between samples from the various companies in terms of spectral patterns and biochemical composition. Differences between samples from the various companies were small compared to the biochemical effects of the toxin. A metabonomic model has been constructed for urine from control rats, enabling identification of outlier samples and the metabolic reasons for the deviation. Building on this success, and with the completion of studies on approximately 80 model toxins, first expert systems for prediction of liver and kidney toxicity have been generated.

Automated mode-of-action detection by metabolic profiling

Rapid classification and identification of the mode-of-action of bioactive compounds applied to plants can be achieved by a robust and easy-to-use metabolic-profiling method. This method uses artificial neural network analysis of one-dimensional proton NMR spectra of aqueous plant extracts to rapidly classify changes in the total metabolic profile caused by application of crop protection chemicals.

Metabonomic characterization of genetic variations in toxicological and metabolic responses using probabilistic neural networks

Current emphasis on efficient screening of novel therapeutic agents in toxicological studies has resulted in the evaluation of novel analytical technologies, including genomic (transcriptomic) and proteomic approaches. We have shown that high-resolution 1H NMR spectroscopy of biofluids and tissues coupled with appropriate chemometric analysis can also provide complementary data for use in in vivo toxicological screening of drugs. Metabonomics concerns the quantitative analysis of the dynamic multiparametric metabolic response of living systems to pathophysiological stimuli or genetic modification [Nicholson, J. K., Lindon, J. C., and Holmes, E. (1999) Xenobiotica 11, 1181-1189]. In this study, we have used 1H NMR spectroscopy to characterize the time-related changes in the urinary metabolite profiles of laboratory rats treated with 13 model toxins and drugs which predominantly target liver or kidney. These 1H NMR spectra were data-reduced and subsequently analyzed using a probabilistic neural network (PNN) approach. The methods encompassed a database of 1310 samples, of which 583 comprised a training set for the neural network, with the remaining 727 (independent cases) employed as a test set for validation. Using these techniques, the 13 classes of toxicity, together with the variations associated with strain, were distinguishable to >90%. Analysis of the 1H NMR spectral data by multilayer perceptron networks and principal components analysis gave a similar but less accurate classification than PNN analysis. This study has highlighted the value of probabilistic neural networks in developing accurate NMR-based metabonomic models for the prediction of xenobiotic-induced toxicity in experimental animals and indicates possible future uses in accelerated drug discovery programs. Furthermore, the sensitivity of this tool to strain differences may prove to be useful in investigating the genetic variation of metabolic responses and for assessing the validity of specific animal models.

An NMR-based metabonomic approach to investigate the biochemical consequences of genetic strain differences: application to the C57BL10J and Alpk:ApfCD mouse

As the human genome sequencing projects near completion, there is an active search for technologies that can provide insights into the genetic basis for physiological variation and interpreting gene expression in terms of phenotype at the whole organism level in order to understand the pathophysiology of disease. We present a novel metabonomic approach to the investigation of genetic influences on metabolic balance and metabolite excretion patterns in two phenotypically normal mouse models (C57BL10J and Alpk:ApfCD). Chemometric techniques were applied to optimise recovery of biochemical information from complex (1)H NMR urine spectra and to determine metabolic biomarker differences between the two strains. Differences were observed in tricarboxylic acid cycle intermediates and methylamine pathway activity. We suggest here a new 'metabotype' concept, which will be of value in relating quantitative physiological and biochemical data to both phenotypic and genetic variation in animals and man.

From magnetic resonance spectroscopy to classification of tumors. A review of pattern recognition methods

This article reviews the wealth of different pattern recognition methods that have been used for magnetic resonance spectroscopy (MRS) based tumor classification. The methods have in common that the entire MR spectra is used to develop linear and non-linear classifiers. The following issues are addressed: (i) pre-processing, such as normalization and digitization, (ii) extraction of relevant spectral features by multivariate methods, such as principal component analysis, linear discriminant analysis (LDA), and optimal discriminant vector, and (iii) classification by LDA, cluster analysis and artificial neural networks. Different approaches are compared and discussed in view of practical and theoretical considerations.

Assessment of statistical and neural networks methods in NMR spectral classification and metabolite selection

Magnetic resonance spectroscopy opens a window into the biochemistry of living tissue. However, spectra acquired from different tissue types in vivo or in vitro and from body fluids contain a large number of peaks from a range of metabolites, whose relative intensities vary substantially and in complicated ways even between successive samples from the same category. The realization of the full clinical potential of NMR spectroscopy relies, in part, on our ability to interpret and quantify the role of individual metabolites in characterizing specific tissue and tissue conditions. This paper addresses the problem of tissue classification by analysing NMR spectra using statistical and neural network methods. It assesses the performance of classification models from a range of statistical methods and compares them with the performance of artificial neural network models. The paper also assesses the consistency of the models in selecting, directly from the spectra, the subsets of metabolites most relevant for differentiating between tissue types. The analysis techniques are examined using in vitro spectra from eight classes of normal tissue and tumours obtained from rats. We show that, for the given data set, the performance of linear and non-linear methods is comparable, possibly due to the small sample size per class. We also show that using a subset of metabolites selected by linear discriminant analysis for further analysis by neural networks improves the classification accuracy, and reduces the number of metabolites necessary for correct classification.

Development of a model for classification of toxin-induced lesions using 1H NMR spectroscopy of urine combined with pattern recognition

Pattern recognition approaches were developed and applied to the classification of 600 MHz 1H NMR spectra of urine from rats dosed with compounds that induced organ-specific damage in either the liver or kidney. Male rats were separated into groups (n = 5) and each treated with one of the following compounds; adriamycin, allyl alcohol, 2-bromoethanamine hydrobromide, hexachlorobutadiene, hydrazine, lead acetate, mercury II chloride, puromycin aminonucleoside, sodium chromate, thioacetamide, 1,1,2-trichloro-3,3,3-trifluoro-1-propene or dose vehicle. Urine samples were collected over a 7 day time-course and analysed using 600 MHz 1H NMR spectroscopy. Each NMR spectrum was data-reduced to provide 256 intensity-related descriptors of the spectra. Data corresponding to the periods 8-24 h, 24-32 h and 32-56 h post-dose were first analysed using principal components analysis (PCA). In addition, samples obtained 120-144 h following the administration of adriamycin and puromycin were included in the analysis in order to compensate for the late onset of glomerular toxicity. Having established that toxin-related clustering behaviour could be detected in the first three principal components (PCs), three-quarters of the data were used to construct a soft independent modelling of class analogy (SIMCA) model. The remainder of the data were used as a test set of the model. Only three out of 61 samples in the test set were misclassified. Finally as a further test of the model, data from the 1H NMR spectra of urine from rats that had been treated with uranyl nitrate were used. Successful prediction of the toxicity type of the compound was achieved based on NMR urinalysis data confirming the robust nature of the derived model.

Application of self-organizing maps for the detection and classification of human blood plasma lipoprotein lipid profiles on the basis of 1H NMR spectroscopy data

Efficient and relevant classification of clinical findings, i.e. diagnostic decision making, poses a major challenge in medicine. In relation to biomedical NMR spectroscopy the problem of classification is often accompanied by complex, heavily overlapping information. Self-organizing map (SOM) analysis has been successfully applied in many areas of research and was thus also considered as a potential tool for NMR data analysis. In this paper we demonstrate how SOM analysis can be used for automated NMR data classification. Our goal was analysis of plasma lipoprotein lipids, a complex but biochemically well understood and specified system. The results illustrate that clinically relevant lipid classifications can be obtained from the SOM analysis of 1H NMR spectral information alone. The resulting maps were calibrated using independent biochemical lipid analyses and were found to produce excellent clustering of the plasma samples into clinically useful groups: normal, type IIa, IIb and IV hyperlipidaemias. In addition to this traditional classification, we also present results from SOM analysis in which the reference vectors of the map were calibrated for plasma total cholesterol and triglycerides and high and low density lipoprotein C; the plasma lipid parameters that are currently considered as the most useful indicators of coronary heart disease risk. In all, the present results indicate that SOM analysis can cope well with complex NMR spectral information and is thus likely to have an independent role in the area of biomedical NMR data analysis.

Renal papillary necrosis--40 years on

Analgesics and nonsteroidal anti-inflammatory drugs (NSAIDs) are well recognized as a major class of therapeutic agent that causes renal papillary necrosis (RPN). Over the last decade a broad spectrum of other therapeutic agents and many chemicals have also been reported that have the potential to cause this lesion in animals and man. There is consensus that RPN is the primary lesion that can progress to cortical degeneration; and it is only at this stage that the lesion is easily diagnosed. In the absence of sensitive and selective noninvasive biomarkers of RPN there is still no clear indication of which compound, under what circumstances, has the greatest potential to cause this lesion in man. Attempts to mimic RPN in rodents using analgesics and NSAIDs have not provided robust models of the lesion. Thus, much of the research has concentrated on those compounds that cause an acute or subacute RPN as the basis by which to study the pathogenesis of the lesion. Based on the mechanistic understanding gleaned from these model compounds it has been possible to transpose an understanding of the underlying processes to the analgesics and NSAIDs. The mechanism of RPN is still controversial. There are data that support microvascular changes and local ischemic injury as the underlying cause. Alternatively, several model papillotoxins, some analgesics, and NSAIDs target selectively for the medullary interstitial cells, which is the earliest reported aberration, after which there are a series of degenerative processes affecting other renal cell types. Many papillotoxins have the potential to undergo prostaglandin hydroperoxidase-mediated metabolic activation, specifically in the renal medullary interstitial cells. These reactive intermediates, in the presence of large quantities of polyunsaturated lipid droplets, result in localized and selective injury of the medullary interstitial cells. These highly differentiated cells do not repair, and it is generally accepted that continuing insult to these cells will result in their progressive erosion. The loss of these cells is thought to be central to the degenerative cascade that affects the cortex. There is still a need to understand better the primary mechanism and the secondary consequences of RPN so that the risk of chemical agents in use and novel molecules can be fully assessed.

Normal ultrastructure of the kidney and lower urinary tract

The mammalian urinary tract includes the kidneys, ureters, urinary bladder, and urethra. The renal parenchyma is composed of the glomeruli and a heterogeneous array of tubule segments that are specialized in both function and structure and are arranged in a specific spatial distribution. The ultrastructure of the glomeruli and renal tubule epithelia have been well characterized and the relationship between the cellular structure and the function of the various components of the kidney have been the subject of intense study by many investigators. The lower urinary tract, the ureters, urinary bladder, and urethra, which are histologically similar throughout, are composed of a mucosal layer lined by transitional epithelium, a tunica muscularis, and a tunica serosa or adventitia. The present manuscript reviews the normal ultrastructural morphology of the kidney and the lower urinary tract. The normal ultrastructure is illustrated using transmission electron microscopy of normal rat kidney and urinary bladder preserved by in vivo perfusion with glutaraldehyde fixative and processed in epoxy resin.

Assessment of quantitative artificial neural network analysis in a metabolically dynamic ex vivo 31P NMR pig liver study

Quantitative artificial neural network analysis for 1550 ex vivo 31P nuclear magnetic resonance spectra from hypothermically reperfused pig livers was assessed. These spectra show wide ranges of metabolite concentrations and have been analyzed using metabolite prior knowledge based lineshape fitting analysis which had proved robust in its biochemical interpretation. This finding provided a good opportunity to assess the performance of artificial neural network analysis in a biochemically complex situation. The results showed high correlations (0.865 < or = R < or = 0.992) between the lineshape fitting and artificial neural network analysis for the metabolite values, and the artificial neural network analysis was able to fully represent the trends in the metabolic fluctuations during the experiments.