Toxicogenomic analysis methods for predictive toxicology

Identification of early liver toxicity gene biomarkers using comparative supervised machine learning

Screening agrochemicals and pharmaceuticals for potential liver toxicity is required for regulatory approval and is an expensive and time-consuming process. The identification and utilization of early exposure gene signatures and robust predictive models in regulatory toxicity testing has the potential to reduce time and costs substantially. In this study, comparative supervised machine learning approaches were applied to the rat liver TG-GATEs dataset to develop feature selection and predictive testing. We identified ten gene biomarkers using three different feature selection methods that predicted liver necrosis with high specificity and selectivity in an independent validation dataset from the Microarray Quality Control (MAQC)-II study. Nine of the ten genes that were selected with the supervised methods are involved in metabolism and detoxification (Car3, Crat, Cyp39a1, Dcd, Lbp, Scly, Slc23a1, and Tkfc) and transcriptional regulation (Ablim3). Several of these genes are also implicated in liver carcinogenesis, including Crat, Car3 and Slc23a1. Our biomarker gene signature provides high statistical accuracy and a manageable number of genes to study as indicators to potentially accelerate toxicity testing based on their ability to induce liver necrosis and, eventually, liver cancer.

Development of a data-processing method based on Bayesian k-means clustering to discriminate aneugens and clastogens in a high-content micronucleus assay

Genotoxicants can be identified as aneugens and clastogens through a micronucleus (MN) assay. The current high-content screening-based MN assays usually discriminate an aneugen from a clastogen based on only one parameter, such as the MN size, intensity, or morphology, which yields low accuracies (70-84%) because each of these parameters may contribute to the results. Therefore, the development of an algorithm that can synthesize high-dimensionality data to attain comparative results is important. To improve the automation and accuracy of detection using the current parameter-based mode of action (MoA), the MN MoA signatures of 20 chemicals were systematically recruited in this study to develop an algorithm. The results of the algorithm showed very good agreement (93.58%) between the prediction and reality, indicating that the proposed algorithm is a validated analytical platform for the rapid and objective acquisition of genotoxic MoA messages.

Characteristic molecular signatures of early exposure to volatile organic compounds in rat liver

OBJECTIVE

Investigation on whether the characteristic molecular signatures can discriminate individual volatile organic compounds (VOCs) and provide predictive markers for the detection of VOC exposure.

METHODS

Transcriptomic analysis of liver tissues was performed 48 h after the single oral administration of three VOCs doses at LD25 or LD5 values, to Sprague-Dawley.

RESULTS

Combination analysis of different multi-classifications suggested that 145 genes predicted VOC exposure. Additionally, Gene Set Enrichment Analysis of genes deregulated by VOCs revealed that T cell prolymphatic leukemia signaling was inactivated in all VOCs.

CONCLUSIONS

These molecular markers could be widely implemented to assess and predict environmental exposure to VOCs.

Characteristic molecular signature for early detection and prediction of persistent organic pollutants in rat liver

Persistent organic pollutants (POPs) are degradation-resistant anthropogenic chemicals that accumulate in the food chain and in adipose tissue, and are among the most hazardous compounds ever synthesized. However, their toxic mechanisms are still undefined. To investigate whether characteristic molecular signatures can discriminate individual POP and provide prediction markers for the early detection of POPs exposure in an animal model, we performed transcriptomic analysis of rat liver tissues after exposure to POPs. The six different POPs (toxaphene, hexachlorobenzene, chlordane, mirex, dieldrin, and heptachlor) were administered to 11-week-old male Sprague-Dawley rats, and after 48 h of exposure, RNAs were extracted from liver tissues and subjected to rat whole genome expression microarrays. Early during exposure, conventional toxicological analysis including changes in the body and organ weight, histopathological examination, and blood biochemical analysis did not reflect any toxicant stresses. However, unsupervised gene expression analysis of rat liver tissues revealed in a characteristic molecular signature for each toxicant, and supervised analysis identified 2708 outlier genes that discerned the POPs exposure group from the vehicle-treated control. Combination analysis of two different multiclassifications suggested 384 genes as early detection markers for predicting each POP exposure with 100% accuracy. The data from large-scale gene expression analysis of a different POP exposure in rat model suggest that characteristic expression profiles exist in liver hepatic cells and multiclassification of POP-specific molecular signatures can discriminate each toxicant at an early exposure time. The use of these molecular markers may be more widely implemented in combination with more traditional techniques for assessment and prediction of toxicity exposure to POPs from an environmental aspect.

Characteristic molecular signature for the early detection and prediction of polycyclic aromatic hydrocarbons in rat liver

Predictions of toxicity are central for the assessment of chemical toxicity, and the effects of environmental toxic compounds are still a major issue for predicting potential human health risks. Among the various environmental toxicants, polycyclic aromatic hydrocarbons (PAHs) are an important class of environmental pollutant, and many PAHs are known or suspected carcinogens. In the present study, to investigate whether characteristic expression profiles of PAHs exist in rat liver and whether a characteristic molecular signature can discriminate and predict among different PAHs at an early exposure time, we analyzed the genome-wide expression profiles of rat livers exposed to PAHs [benzo[a]anthracene (BA), benzo[a]pyrene (BP), phenanthrene (PA) and naphthalene (NT)]. At early time-point PAH exposure, large-scale gene expression analysis resulted in characteristic molecular signatures for each PAH, and supervised analysis identified 1183 outlier genes as a distinct molecular signature discerning PAHs from the normal control group. We identified 158 outlier genes as early predictive and surrogate markers for predicting each tested PAH by combination of two different multi-classification algorithms with 100% accuracy through a leave-one out cross-validation method. In conclusion, the characteristic gene expression signatures from a rat model system could be used as predictable and discernible gene-based biomarkers for the detection and prediction of PAHs, and these molecular markers may provide insights into the underlying mechanisms for genotoxicity of exposure to PAHs from environmental aspect.

Biomarker classifiers for identifying susceptible subpopulations for treatment decisions

AIM

A main goal of pharmacogenomics is to develop genomic signatures to predict patients' responses to a drug or therapy for treatment decisions. Identification of patients who would have no beneficial effect or have the risk of developing adverse effects from an unnecessary treatment could save enormous cost in the healthcare system and clinical trials. This article presents an approach for developing a biomarker classifier for identifying a fraction of susceptible patients, who should be spared unnecessary treatment prior to treatment.

MATERIALS & METHODS

The identification of susceptible patients involves two steps. The first step is to identify biomarkers of susceptibility from a mixture of biomarkers of susceptibility and biomarkers of response; the second step is to develop a classifier using an ensemble classification algorithm, as the number of susceptible patients is generally much smaller than the number of nonsusceptible patients.

RESULTS

Selection of the biomarkers of susceptibility is essential to achieve good prediction accuracy. The ensemble algorithm significantly improves the prediction accuracy compared with the standard classifiers.

CONCLUSION

The study shows that classifiers developed based on the biomarkers obtained by comparing the genomic profiles of responders to those of nonresponders may lead to a high misclassification error rate. Classifiers to identify a small fraction of the subpopulation should take imbalanced class sizes into consideration. A large sample size may be needed in order to ensure detection of a sufficient number of biomarkers and a sufficient number of susceptible subjects for classifier development and validation.

An approach to identifying preclinical biomarkers of susceptibility to drug-induced toxicity

Aim: Drug-induced toxicity that leads to termination of candidate drugs or postmarketing removal of approved drugs can potentially be explained by the existence of susceptible subpopulations. If the susceptible subpopulations are identified in advance, the drug’s benefits could be maximized by optimal treatment decisions. This article presents a statistical model and an approach for identifying pharmacogenomic biomarkers of susceptibility to drug-induced toxicity for detecting the susceptible subpopulations. Materials & methods: Biomarkers are categorized into three disjoint sets: biomarkers of both susceptibility and exposure (A); biomarkers of susceptibility only (B); and biomarkers of exposure only (C). Set B contains the most useful biomarkers to identify susceptible subpopulations prior to drug exposure; these markers demonstrate no change in response before and after drug exposure. We present a sample size analysis to illustrate the issues and challenges facing identifying biomarker set B. Results: The required sample size increases as the proportion of the susceptible subpopulation decreases. The examples demonstrated that at least 75 subjects per group are needed for a population with a 5% susceptible subpopulation and more than 1000 are often necessary. Conclusion: This study demonstrates that the biomarkers identified by common methods are a mixture of biomarkers of exposure and susceptibility (A and C), it further demonstrates that the proposed approach could be used to identify biomarkers of susceptibility (B), where a large sample size may be required for adequate power and low false-positive rate.

Original submitted 14 October 2010; Revision submitted 8 December 2010.

Molecular signature for early detection and prediction of polycyclic aromatic hydrocarbons in peripheral blood

Whole blood is one of the most easily accessible biofluids, and circulating leukocytes would include informative transcripts as a first line of immune defense for many disease processes. To demonstrate that transcriptomic responses of circulating blood cells reflect the exposure to environmental toxicants and the characteristic molecular signatures can discriminate and predict the type of toxicant at an early exposure time, we identified and validated characteristic gene expression profiles of rat whole blood after exposure to polycyclic aromatic hydrocarbons (PAHs). At an early exposure time point, conventional toxicological analysis including changes in the body and organ weight, histopathological examination, and blood biochemical analysis did not reflect any toxicant stresses. However, unsupervised gene expression analysis of blood cells resulted in a characteristic molecular signature for each toxicant. Further analysis of multiclassification suggested 220 genes as early detective and surrogate markers for predicting each PAH with 100% accuracy. These findings suggest that the blood expression signature could be used as a predictable and discernible surrogate marker for detection and prediction of PAHs, and the use of these molecular markers may be more widely implemented in combination with more traditional techniques for assessment and prediction of toxicity exposure to PAHs from an environmental aspect.

Identification of characteristic molecular signature for volatile organic compounds in peripheral blood of rat

In a previous report we demonstrated that the transcriptomic response of liver tissue was specific to toxicants, and a characteristic molecular signature could be used as an early prognostic biomarker in rats. It is necessary to determine the transcriptomic response to toxicants in peripheral blood for application to the human system. Volatile organic compounds (VOCs) comprise a major group of pollutants which significantly affect the chemistry of the atmosphere and human health. In this study we identified and validated the specific molecular signatures of toxicants in rat whole blood as early predictors of environmental toxicants. VOCs (dichloromethane, ethylbenzene, and trichloroethylene) were administered to 11-week-old SD male rats after 48h of exposure, peripheral whole blood was subjected to expression profiling analysis. Unsupervised gene expression analysis resulted in a characteristic molecular signature for each toxicant, and supervised analysis identified 1,217 outlier genes as distinct molecular signatures discerning VOC exposure from healthy controls. Further analysis of multi-classification suggested 337 genes as early detective molecular markers for three VOCs with 100% accuracy. A large-scale gene expression analysis of a different VOC exposure animal model suggested that characteristic expression profiles exist in blood cells and multi-classification of this VOC-specific molecular signature can discriminate each toxicant at an early exposure time. This blood expression signature can thus be used as discernable surrogate marker for detection of biological responses to VOC exposure in an environment.

Use of SELDI MS to discover and identify potential biomarkers of toxicity in InnoMed PredTox: a multi-site, multi-compound study

A serious bottleneck in the drug development pipeline is the inability of current pre-clinical toxicology evaluation methods to predict early on, and with good accuracy, that a drug candidate will have to be removed from development due to toxicology/safety issues. The InnoMed PredTox consortium attempted to address this issue by assessing the value of using molecular profiling techniques (proteomics, transcriptomics, and metabonomics), in combination with conventional toxicology measurements, on decision making earlier in pre-clinical safety evaluation. In this study, we report on the SELDI-TOF-MS proteomics component of the InnoMed PredTox project. In this large scale, multi-site, multi-compound study, tissue and plasma samples from 14-day in vivo rat experiments conducted for 16 hepato- and nephro-toxicants with known toxicology endpoints (including 14 proprietary compounds and 2 reference compounds) were analyzed by SELDI-TOF-MS. We have identified seven plasma proteins and four liver proteins which were shown to be modulated by treatment, and correlated with histopathological evaluations and can be considered potential biomarker candidates for the given toxicology endpoints. In addition, we report on the intra- and inter-site variations observed based on measurements from a reference sample, and steps that can be taken to minimize this variation.

Whole genomic expression analysis of octachlorostyrene-induced chronic toxicity in Caenorhabditis elegans

In recent years, microarray technology has enabled the investigation of possible mechanisms the expression of genes related to toxic compounds. We used a C. elegans whole genome microarray to observe and evaluate the chronic toxicity of the free-living nematode Caenorhabditis elegans (C. elegans) after exposure to octachlorostyrene, (OCS), a by-product in the manufacture of many chlorinated hydrocarbons. In this study, we examined sublethal toxicity, egg hatching, and movement of octachlorostyrene over three generations using a nematode growth medium (NGM) agar plate. In the third generation, OCS affected the fecundity rate of C. elegans. Specifically, the number of worm and eggs decreased significantly to about 50% of control (p < 0.05). In microarray experiments, total RNA was isolated at 0, 2 and 3 generations following treatment of OCS, and hybridized to the microarray containing about 22,000 C. elegans genes. Dye swaps were performed. After data analysis, we identified a total of 1,294 genes that were differentially expressed through generations.

Expression analysis of early response-related genes in rat liver epithelial cells exposed to thioacetamide in vitro

Thioacetamide (TA) is a potent hepatotoxicant known to affect liver metabolism, inhibit mRNA transport and induce immune suppression. The genetic mechanism underlining this biological toxic compound is well understood using microarray technology. Thus, we used high-throughput rat genome oligonucleotide microarrays containing approximately 22,000 genes to investigate the genetic components of TA-related cytotoxicity in WB-F344 rat liver epithelial (WB-F344) cells. We treated cells with TA (two concentrations over five time periods, ranging from 1 to 24 hr), isolated total RNA at 1, 3, 6, 12 and 24 hr following TA treatment and hybridized the RNA to microarrays. Clustering analysis distinguished two groups of genes, early (1 and 3 hr) and late (6, 12 and 24 hr) phase genes. In total, 2,129 and 2,348 differentially-expressed genes were identified following treatment with low and high concentrations of TA, respectively. A common set of 1,229 genes that were differentially expressed following treatment with both low (1,000 muM) and high (10,000 muM) concentrations of TA had similar expression patterns. Interestingly, 1,410 genes at the low concentration and 1,858 genes at the high concentration were differentially expressed in the early phases, suggesting that these genes associated with the early response to TA may be useful as early markers of hepatotoxicity.

Toxicogenomics: transcription profiling for toxicology assessment

Toxicogenomics, the application of transcription profiling to toxicology, has been widely used for elucidating the molecular and cellular actions of chemicals and other environmental stressors on biological systems, predicting toxicity before any functional damages, and classification of known or new toxicants based on signatures of gene expression. The success of a toxicogenomics study depends upon close collaboration among experts in different fields, including a toxicologist or biologist, a bioinformatician, statistician, physician and, sometimes, mathematician. This review is focused on toxicogenomics studies, including transcription profiling technology, experimental design, significant gene extraction, toxicological results interpretation, potential pathway identification, database input and the applications of toxicogenomics in various fields of toxicological study.

Acute Hepatotoxicity: A Predictive Model Based on Focused Illumina Microarrays

Drug-induced hepatotoxicity is a major issue for drug development, and toxicogenomics has the potential to predict toxicity during early toxicity screening. A bead-based Illumina oligonucleotide microarray containing 550 liver specific genes has been developed. We have established a predictive screening system for acute hepatotoxicity by analyzing differential gene expression profiles of well-known hepatotoxic and nonhepatotoxic compounds. Low and high doses of tetracycline, carbon tetrachloride (CCL4), 1-naphthylisothiocyanate (ANIT), erythromycin estolate, acetaminophen (AAP), or chloroform as hepatotoxicants, clofibrate, theophylline, naloxone, estradiol, quinidine, or dexamethasone as nonhepatotoxic compounds, were administered as a single dose to male Sprague-Dawley rats. After 6, 24, and 72 h, livers were taken for histopathological evaluation and for analysis of gene expression alterations. All hepatotoxic compounds tested generated individual gene expression profiles. Based on leave-one-out cross-validation analysis, gene expression profiling allowed the accurate discrimination of all model compounds, 24 h after high dose treatment. Even during the regeneration phase, 72 h after treatment, CCL4, ANIT, and AAP were predicted to be hepatotoxic, and only these three compounds showed histopathological changes at this time. Furthermore, we identified 64 potential marker genes responsible for class prediction, which reflected typical hepatotoxicity responses. These genes and pathways, commonly deregulated by hepatotoxicants, may be indicative of the early characterization of hepatotoxicity and possibly predictive of later hepatotoxicity onset. Two unknown test compounds were used for prevalidating the screening test system, with both being correctly predicted. We conclude that focused gene microarrays are sufficient to classify compounds with respect to toxicity prediction.

Toxicogenomics of non-viral drug delivery systems for RNAi: potential impact on siRNA-mediated gene silencing activity and specificity

RNA interference (RNAi) is an evolutionary conserved cellular process for the regulation of gene expression. In mammalian cells, RNAi is induced via short (21-23 nt) duplexes of RNA, termed small interfering RNA (siRNA), that can elicit highly sequence-specific gene silencing. However, synthetic siRNA duplexes are polyanionic macromolecules that do not readily enter cells and typically require the use of a delivery vector for effective gene silencing in vitro and in vivo. Choice of delivery system is usually made on its ability to enhance cellular uptake of siRNA. However, recent gene expression profiling (toxicogenomics) studies have shown that separate from their effects on cellular uptake, delivery systems can also elicit wide ranging gene changes in target cells that may impact on the 'off-target' effects of siRNA. Furthermore, if delivery systems also alter the expression of genes targeted for silencing, then siRNA activity may be compromised or enhanced depending on whether the target gene is up-regulated or down-regulated respectively. Citing recent examples from the literature, this article therefore reviews the toxicogenomics of non-viral delivery systems and highlights the importance of understanding the genomic signature of siRNA delivery reagents in terms of their impact on gene silencing activity and specificity. Such information will be essential in the selection of optimally acting siRNA-delivery system combinations for the many applications of RNA interference.

Toxicogenomic Study of Triazole Fungicides and Perfluoroalkyl Acids in Rat Livers Predicts Toxicity and Categorizes Chemicals Based on Mechanisms of Toxicity

Toxicogenomic analysis of five environmental chemicals was performed to investigate the ability of genomics to predict toxicity, categorize chemicals, and elucidate mechanisms of toxicity. Three triazole antifungals (myclobutanil, propiconazole, and triadimefon) and two perfluorinated chemicals [perfluorooctanoic acid (PFOA) and perfluorooctane sulfonate (PFOS)] were administered daily via oral gavage for one, three, or five consecutive days to male Sprague-Dawley rats at single doses of 300, 300, 175, 20, or 10 mg/kg/day, respectively. Clinical chemistry, hematology, and histopathology were measured at all time points. Gene expression profiling of livers from three rats per treatment group at all time points was performed on the CodeLink Uniset Rat I Expression array. Data were analyzed in the context of a large reference toxicogenomic database containing gene expression profiles for over 630 chemicals. Genomic signatures predicting hepatomegaly and hepatic injury preceded those results for all five chemicals, and further analysis segregated chemicals into two distinct classes. The triazoles caused similar gene expression changes as other azole antifungals, particularly the induction of pregnane X receptor (PXR)-regulated xenobiotic metabolism and oxidative stress genes. In contrast, PFOA and PFOS exhibited peroxisome proliferator-activated receptor alpha agonist-like effects on genes associated with fatty acid homeostasis. PFOA and PFOS also resulted in downregulation of cholesterol biosynthesis genes, matching an in vivo decrease in serum cholesterol, and perturbation of thyroid hormone metabolism genes matched by serum thyroid hormone depletion in vivo. The concordance of in vivo observations and gene expression findings demonstrated the ability of genomics to accurately categorize chemicals, identify toxic mechanisms of action, and predict subsequent pathological responses.

Quantitative proteomic analysis of rat liver for carcinogenicity prediction in a 28-day repeated dose study

The potential of quantitative proteomic analysis to predict carcinogenicity of chemical compounds was investigated. Using 2D-DIGE, we analyzed the effects of 63 chemical compounds on protein expression in the rat liver after 28 daily doses. Types of carcinogens were categorized depending on the species and organ specificity. The carcinogen characteristic proteins for each classification were identified by Welch's t value. For evaluation of the predictive concordance we used support vector machines. The rat hepatic carcinogen-specific classification gave higher concordance than the other classification. The generalization performance was measured by leave-one-out cross-validation. For genotoxic and non-genotoxic compounds, a concordance of 79.3 and 76.5%, respectively, was obtained by the top 30 ranked proteins with Welch's t value. Furthermore, we found that the increase of the expression level of the stress response proteins as the common feature of poorly predicted chemical compounds in the leave-20%-out cross-validation. Quantitative proteomics could be promising technique for developing biomarker panels that can be used for carcinogenicity prediction. The list of proteins identified in this study and the zoomed gel images of the top ranked proteins in statistic analysis are provided in Supplementary Data.

A dolphin peripheral blood leukocyte cDNA microarray for studies of immune function and stress reactions

A microarray focused on stress response and immune function genes of the bottlenosed dolphin has been developed. Random expressed sequence tags (ESTs) were isolated and sequenced from two dolphin peripheral blood leukocyte (PBL) cDNA libraries biased towards T- and B-cell gene expression by stimulation with IL-2 and LPS, respectively. A total of 2784 clones were sequenced and contig analysis yielded 1343 unigenes (archived and annotated at ). In addition, 52 dolphin genes known to be important in innate and adaptive immune function and stress responses of terrestrial mammals were specifically targeted, cloned and added to the unigene collection. The set of dolphin sequences printed on a cDNA microarray comprised the 1343 unigenes, the 52 targeted genes and 2305 randomly selected (but unsequenced) EST clones. This set was printed in duplicate spots, side by side, and in two replicates per slide, such that the total number of features per microarray slide was 19,200, including controls. The dolphin arrays were validated and transcriptomic profiles were generated using PBL from a wild dolphin, a captive dolphin and dolphin skin cells. The results demonstrate that the array is a reproducible and informative tool for assessing differential gene expression in dolphin PBL and in other tissues.