Comparison of supervised clustering methods to discriminate genotoxic from non-genotoxic carcinogens by gene expression profiling.

Prediction of the toxic properties of chemicals based on modulation of gene expression profiles in exposed cells or animals is one of the major applications of toxicogenomics. Previously, we demonstrated that by Pearson correlation analysis of gene expression profiles from treated HepG2 cells it is possible to correctly discriminate and predict genotoxic from non-genotoxic carcinogens. Since to date many different supervised clustering methods for discrimination and prediction tests are available, we investigated whether application of the methods provided by the Whitehead Institute and Stanford University improved our initial prediction. Four different supervised clustering methods were applied for this comparison, namely Pearson correlation analysis (Pearson), nearest shrunken centroids analysis (NSC), K-nearest neighbour analysis (KNN) and Weighted voting (WV). For each supervised clustering method, three different approaches were followed: (1) using all the data points for all treatments, (2) exclusion of the samples with marginally affected gene expression profiles and (3) filtering out the gene expression signals that were hardly altered. On the complete data set, NSC, KNN and WV outperformed the Pearson test, but on the reduced data sets no clear difference was observed. Exclusion of samples with marginally affected profiles improved the prediction by all methods. For the various prediction models, gene sets of different compositions were selected; in these 27 genes appeared three times or more. These 27 genes are involved in many different biological processes and molecular functions, such as apoptosis, cell cycle control, regulation of transcription, and transporter activity, many of them related to the carcinogenic process. One gene, BAX, was selected in all 10 models, while ZFP36 was selected in 9, and AHR, MT1E and TTR in 8. Summarising, this study demonstrates that several supervised clustering methods can be used to discriminate certain genotoxic from non-genotoxic carcinogens by gene expression profiling in vitro in HepG2 cells. None of the methods clearly outperforms the others.

Discrimination of genotoxic from non-genotoxic carcinogens by gene expression profiling.

Two general mechanisms are implicated in chemical carcinogenesis. The first involves direct damage to DNA, referred to as genotoxic (GTX), to which the cell responds by repair of the damages, arrest of the cell cycle or induction of apoptosis. The second is non-DNA damaging, non-genotoxic (NGTX), in which a wide variety of cellular processes may be involved. Therefore, it can be hypothesized that modulation of the underlying gene expression patterns is profoundly distinct between GTX and NGTX carcinogens, and thus that expression profiling is applicable for classification of chemical carcinogens as GTX or NGTX. We investigated this hypothesis by analysing modulation of gene expression profiles induced by 20 chemical carcinogens in HepG2 cells with application of cDNA microarrays that contain 597 toxicologically relevant genes. In total, 22 treatments were included, divided in two sets. The training set consisted of 16 treatments (nine genotoxins and seven non-genotoxins) and the validation set of six treatments (three and three). Class discrimination models based on Pearson correlation analyses for the 20 most discriminating genes were developed with data from the training set, where after the models were tested with all data. Using all data, the correctness for classification of the carcinogens from the training set was clearly better than that for the validation set, namely 81 and 33%, respectively. Exclusion of the treatments that had only marginal effects on the expression profiles, improved the discrimination for the training and validation sets to 92 and 100% correctness, respectively. Exclusion of the gene expression signals that were hardly altered also improved classification, namely to 94 and 80%. Therefore, our study proves the principle that gene expression profiling can discriminate carcinogens with major differences in their mode of actions, namely genotoxins versus non-genotoxins.

A multi-biomarker approach to study the effects of smoking on oxidative DNA damage and repair and antioxidative defense mechanisms.

We investigated the effects of smoking-induced oxidative stress in healthy volunteers (21 smokers versus 24 non-smokers) by quantifying various markers of oxidative DNA damage and repair, and antioxidative defense mechanisms. Lymphocytic 7-hydroxy-8-oxo-2'-deoxyguanosine (8-oxo-dG) levels measured by high performance liquid chromatography with electrochemical detection, were significantly lower in smokers as compared with non-smokers (38.6 +/- 5.2 versus 50.9 +/- 4.6/10(6) dG, P = 0.05). The levels of oxidized pyrimidine bases in lymphocytes of smokers quantified by the endonuclease III-modified comet assay were non-significantly lower than those of non-smokers (% DNA in tail: 13 +/- 3 versus 14 +/- 2; tail length: 69 +/- 13 versus 96 +/- 10; tail moment: 6416 +/- 1220 versus 7545 +/- 1234). Urinary excretion levels of 8-hydroxy-2'-deoxyguanosine (8-OH-dG) assessed by enzyme-linked immunosorbent assay did not differ significantly between smokers and non-smokers (197 +/- 31 versus 240 +/- 33 ng/body mass index, P = 0.3). Overall DNA repair activity expressed as unscheduled DNA synthesis in blood leukocytes, was not significantly different between smokers and non-smokers (2.9 +/- 0.3 versus 3.3 +/- 0.3, P = 0.4). Plasma antioxidative capacity measured by the Trolox equivalent antioxidant capacity assay was slightly higher in smokers as compared with non-smokers (440 +/- 16 versus 400 +/- 15 microM Trolox equivalent, P = 0.09), and it was significantly related to lymphocytic 8-oxo-dG levels (r = 0.4, P = 0.001). Genotyping of human 8-OH-dG glycosylase/apurinic lyase and glutathione S-transferase M1 showed that a polymorphism in either or both of the two genes does not affect any of the quantified biomarkers. We conclude that oxidative stress imposed by cigarette smoking has a low impact upon certain pathways involved in DNA damage and the antioxidative defense system.

The use of 4-aminobiphenyl hemoglobin adducts and aromatic DNA adducts in lymphocytes of smokers as biomarkers of exposure.

Two biomarkers of exposure to cigarette smoke, 4-aminobiphenyl-hemoglobin (Hb) adducts and aromatic DNA adducts in lymphocytes, were determined from a population of 55 smokers and 4 nonsmokers. The levels of these adducts were related to daily cigarette consumption and also to (calculated) tar and nicotine intake. The Hb adduct levels seemed to correspond best to the number of cigarettes (cig) smoked, but at a cigarette consumption of >30 cig/day, a saturation effect was observed. Lymphocytic DNA adducts also correlated well with cigarette and tar consumption; for this type of adduct, a saturation level was reached at a dose of approximately 15-20 cig/day. From a subpopulation, a second sample was obtained after 2 months, and the adduct levels were compared with their initial adduct levels. Strong correlations were found between the first and second DNA adduct measurements (r = 0.84). In another subpopulation, resampling was performed after 6 months. No correlation between DNA adduct levels in the first and last samples was found, but 4-aminobiphenyl Hb adduct levels were strongly correlated (r = 0.78), the absolute quantities measured being comparable (paired t test: t = -1.27, P = 0.22, n = 15). We found no influence of GSTM1 and NAT2 polymorphisms on Hb adduct formation and of GSTM1 polymorphism on aromatic DNA adduct formation. A significantly lower aromatic DNA adduct level was observed for intermediate acetylators when compared to slow acetylators.