ToxSci Advance Access originally published online on October 20, 2007
Toxicological Sciences 2008 101(2):206-214; doi:10.1093/toxsci/kfm262
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Mode of Action Clustering of Chemicals and Environmental Samples on the Bases of Bacterial Stress Gene Inductions
* University of Antwerp, Department of Biology, Ecophysiology, Biochemistry and Toxicology Group, B-2020 Antwerp, Belgium
VITO, Environmental Toxicology Group, B-2400 Mol, Belgium
1 To whom correspondence should be addressed at University of Antwerp, Department of Biology, Ecophysiology, Biochemistry and Toxicology Group, Groenenborgerlaan 171/U7, B-2020 Antwerp, Belgium. Fax: +32-32-65-04-97. E-mail: gosia.freddy{at}scarlet.be.
Received August 7, 2007; accepted September 26, 2007
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Over the years, environment and the human population have seen an increasing exposure to both existing and newly developed chemicals. It is generally accepted that at least some of those are toxic, albeit as pure compound or in combination with others. In response to a growing public awareness and scientific data, the new European chemicals legislation (Registration, Evaluation and Authorization of Chemicals) is under implementation at the moment. As a consequence, during the coming years about 30,000 chemicals have to be assessed on their potential hazard for man and biota. Part of this assessment will be done using existing and new in vitro tests offering insight into the toxicity of chemicals and into their toxicological mode of action. This study presents data on a battery of 14 bacterial reporter gene assay allowing mode of action determination and statistical grouping of chemicals based on their induction profile. Gene induction results are used to group reference chemicals in a statistical cascade employing hierarchical tree and k-means clustering for initial grouping. Both complementary, yet mathematically different, algorithms are consequently confirmed by principal component analysis (PCA). The gene induction profiles of an environmental extract with documented in vivo effects and a chemical with limited toxicological are data available and projected in the PCA vector space. The projection allows correct mode of action grouping and indicates that effect predictions based on the known toxicological effects of the reference compounds can be made.
Key Words: alternatives to animal testing; in vitro and alternatives; biomonitoring; risk assessment; dose–response; risk assessment; gene expression/regulation; predictive toxicology; toxicogenomics; methods.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Over the last decades, a range of newly developed chemicals have found their way to the market and hence to the (human) environment. Many newly developed chemicals may potentially be toxic to humans and/or the environment. Increased public and scientific awareness has led to a growing claim for sound toxicological data and to the implementation of new regulatory demands. Hence, the new European chemicals legislation (REACH, Registration, Evaluation and Authorization of Chemicals) is under implementation at the moment. It may be noted that the higher number of 30,000 chemicals to be tested is due to the testing of existing chemicals, for which no data have been provided so far. Projections on how many animals would have to be sacrificed due to REACH differ. If standard laboratory animal experiments were to be used, the additional use of 7.5 milj. animals has been estimated (Hofer et al., 2004
). The latter figure takes into account reduction effects such as post hoc acceptance of historical non-OECD compliant data. The amount of testing required for compliance with the REACH guideline far surpasses the capacity of routine screening labs today, and hence it is clear that (at least partial) replacement of animal testing by alternative in vitro assays is not just an option, yet a prerequisite to achieve the goals set by REACH (Combes et al., 2003). A possible approach would be to collect mechanistic data of a compound as early as possible, allowing for more accurate extrapolation toward animal and human toxicity, and using these data to prevent redundant animal testing (Waring and Ulrich, 2000).
Anthropogenic discharges into the environment often consist of complex mixtures of chemicals, many of which are not routinely monitored or detected by chemical analysis. Hence, very often the burden of chemical contamination remains ill defined because chemical analysis can only detect a limited number of compounds. Ecotoxicology has countered this by implementing various biological effect measurements on both whole organism (Adams et al., 2002; Bervoets and Blust, 2003) and in vitro test systems (Dardenne et al., 2007b; Houtman et al., 2004; Johnson and Long, 1998). Recently, the balance is shifting toward an increasing use of in vitro assays, as shown by the growing amount of scientific literature on the subject. Among the in vitro assays, gene induction/repression assays, respectively, based on the on or offset of reporter genes in bacterial or eukaryotic transgenic cells are increasingly used (Harms et al., 2006; Kohler et al., 2000; Sorensen et al., 2006).
Most in vitro assays are fast, fairly simple to perform and interpret; they are of reasonably small scale, cost effective, and mostly at least partially automatable (Eggen and Segner, 2003; Sorensen et al., 2006). Bioassays based on the induction/repression of specific signaling pathways in the living cell are able to combine the detection of toxic compounds (Houtman et al., 2004; White et al., 1997), with the unraveling of their toxic mechanism (Dardenne et al., 2007a; Reifferscheid and Hell, 1996). Understanding the mode of action or mechanisms of toxicity is an important advantage in environmental and human hazard and risk assessment of toxicants, and the extrapolation to higher effect levels (Escher and Hermens, 2002) is necessary to make appropriate predictions of the real impact.
This study explores to what extent compounds can be grouped according to their toxicological mechanism of action using a multiendpoint bacterial gene-profiling test in combination with standard statistical exploratory techniques, that is, cluster and principal component analysis (PCA). On the basis of the gene induction profile obtained from 23 known chemicals, a three-phased approach was used. In a first stage using hierarchical tree clustering, different distance measures and linkage methods were evaluated for their capacity to develop the most logical tree as compared with the known mode of action of the reference compounds (RC). In the second phase, k-means clustering was used on the same data set as an inherently different technique to objectively ascertain the composition of the clusters previously obtained. Third, a PCA was performed to obtain a vector space allowing graphical output combined with objective positioning of compounds with unknown mode of action or complex environmental samples (or extracts thereof) among their closest neighbors. The validity of this approach is then challenged using gene-profiling data from a model compound (CGS 15943) not included in the reference set, and a thoroughly studied pore water sample from a small river basin in Flanders, Belgium. Higher-level effect data obtained from the pore water sample during a prior field study and data available in literature on CGS 15943 are compared with effects induced at higher levels of biological organization by the RC mapping adjacent to the pore water sample and CGS 15943 as shown by PCA.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Bacterial Strains and Bacterial Gene-Profiling Assay
The bacterial strains, except PQ37, and the procedure for the bacterial gene-profiling assay are described elsewhere (Dardenne et al., 2007a
). The PQ37 strain is described in White et al. (1996).
All strains were cultured in Luria Bertani Broth according to standard protocols and kept frozen in 15% (vol/vol) glycerol at –80°C prior to use (Sambrook et al., 1989). All chemicals were purchased at UCB, Belgium and of proanalysis grade, unless specified in the text. The bacteria were exposed at the onset of exponential growth for 90 min, after which cells were lysed and the β-galactosidase activity was determined.
Calculations.
The fold induction at any given dose i for every strain is calculated as the ratio between the β-galactosidase activity at dose i and the average β-galactosidase activity of the nonexposed cells (see Equation 1). The expression in both denominators (measurements at 420 nm) reflects the β-galactosidase activity, whereas the expressions in the nominator (measurements at 600 nm) are a correction factor for the amount of cells present in the assay during the exposure phase, that is, the surface under the growth curve during exposure assuming exponential growth. The predose (PD) optical density (OD) is the OD right before adding the compound to the cells, and the OD at the start of exposure (SE) is the OD measured exactly after dosing the compound. The difference between PD and SE corrects for possible color change as a result of compound addition to the cells.
 | (1) |
Reference Compounds and Tester Samples
RC were chosen from an in house library of approximately 200 profiles based on their theoretical classification in terms of mechanism of action and corresponding gene induction profile (Table 2 in Supplementary Data). Care was taken that different modes of actions were included in the compound set. The RC represent chemicals with distinct gene induction profiles, for example, clear-cut DNA-damaging agents such as methyl methane sulfonate, as well as compounds with more general toxicological features such as methanol. All chemicals used were of at least analytical quality. Stock solutions of model toxicants were made up in ultrapure MilliQ water, except for pentachlorophenol (ethanol p.a.) and toxaphene and 4-nitroquinoline-N-oxide (DMSO). The latter three exposures contained 5% solvent (actual exposure concentration), as did the respective nonexposed controls. All RC were supplied by Sigma-Aldrich (St Louis, MO), all other chemicals by Merck (Germany) or UCB (Belgium) unless otherwise stated. All exposures were performed in triplicate on a 1/2 dilution series from the top concentration down. In total, seven doses and three solvent-only exposures were tested per replicate, giving three replicates per dose and nine solvent-only exposures per compound.
|
Two tester samples (Table 2) were used. The sample designated location 3 C18, is a C18 pore water extract taken from a small effluent dominated river basin in Flanders, Belgium. The sample has a clear industrial fingerprint, yet obeys Belgian and European environmental quality standards, bacterial gene induction profiles and effects at higher levels of biological organization in several tester species were described in detail elsewhere (Dardenne et al., 2007b; Smolders et al., 2004
). CGS 15943 (1,2,4-triazolo-[1,5-c]-quinazolone) is a nonxanthine adenosine receptor antagonist causing behavioral effects that are similar to those produced by classic stimulant drugs. The molecule has antibiotic, antifungal, and antiprotozoan activity and both the core molecule and its substituted derivatives are of interest for their potential anti-inflammatory characteristics (Gineinah et al., 2000; Jantova et al., 2004; Pfeiffer et al., 1999).
Data Analysis
Statistical calculations and significance.
General calculations were done in Excel 2002 (Microsoft Belgium, Diegem, Belgium). Statistical calculations were performed in Statistica version 6 (2001 StatSoft, Inc., Tulsa, OK).
Fold inductions were considered significant (see also Dardenne et al., 2007a,b) based on the following criteria: (1) presence of a dose–response relationship (R2 > 0.50, significant at p < 0.05 for six degrees of freedom) and a positive slope different from 0 (p < 0.05) in a linear model; and (2) signal different from and higher as the blank confirmed by Dunnett's test (p < 0.05). All assays are performed in triplicate on seven doses (1/2 dilution series) and three blanks giving 21 dosed data points and nine blanks per dose–response.
The SfiA marker information is omitted from the dendrogram and PCA calculations as it contains four missing values. Although the results on SfiA (i.e., SOS chromotest) are published (Aflatoxin B1 is positive while ethanol, methanol and menadione are negative), the data were not measured on our assay and thus not listed.
Data Reduction
In order in to standardize the raw data and to reduce the noise before performing the clustering and PCA algorithms, a number of data transformations were carried out. (1) All calculations necessary for clustering and PCA were done on the top dose exposures only. Although the calculations can be done on the fold inductions at lower dose (if still significantly different from the blank), the top dose fold inductions have the better signal to noise ratio and hence give the clearest resolution. (2) All nonsignificant FIs, as determined by the algorithm above, were set to the blank value of 1. (3) Within the range of available promoters, merR takes a special place. Although there are indications that it could slightly react to some forms of DNA damage, it is extremely and specifically triggered by the presence of heavy metal ions, mainly Cd2+ and Hg2+, showing FIs up to a few 100-fold the blank signal. Hence, in the case of high merR inductions all other occurring FIs were considered side effects, and set equal to the blank in order to simplify classification. (4) FI scores (FIS) were calculated as the ratio of the measured FI to the maximum FI (set to 100%, Table 1 in Supplementary Data) attained on approximately 200 compounds and mixtures measured on the same stress promoters. These FIS are used as a cut off value to determine FI values equal to or greater than the 5, 10, and 15% effect levels (EL5, EL10, and EL15). (5) Three extra markers (oxidative stress, general cell lesion, and DNA damage) are calculated as the average of FIS of the promoters belonging to that specific lesion as indicated in Table 1.
|
Clustering Approach
Hierarchical tree (dendrogram) clustering.
Dendrogram clustering is a statistical exploratory technique facilitating the interpretation of the degree of similarity between different multivariate data sets. Different distance (or similarity) measures and linkage algorithms can be explored, and there is no a priori way to determine the best-suited combination. Because we have a good a priori expectation about the structure of the dendrogram we tried all possible combinations attempting to attain the most logical tree. The most consistent results, reflecting best the known mode of action of the RC panel, were obtained using complete linkage (furthest neighbor) in combination with Manhattan distance (taxicab or city block distance). Complete linkage defines distance as the maximum distance between an observation in one cluster and an observation in the other cluster; thus, this method ensures that all observations in a cluster are within that maximum distance. Manhattan distance is the distance between two points measured along axes at right angles, that is, the multidimensional equivalent of coordinates in two-dimensional orthogonal axes. In this measure, the effect of outliers is dampened. The dendrograms of other linkage-distance method combinations are not shown. The combination of complete linkage with the Manhattan distance will be used to determine whether the raw FI values, FIS, or effect levels (EL5, EL10, or EL15) provide the dendrogram that best reflects the expectations based on the mode of action.
k-means clustering.
k-means clustering is different from hierarchical tree clustering as it starts from a beforehand defined number of clusters (k). k-means works toward a clustering of observations in these k groups such that the within-cluster variance is minimized. The algorithm is iterative and may be sensitive to the choice of starter centers. To reduce this effect, starting centers were placed by ranking, all cases by their initial distances, and putting centers at constant intervals.
Principal component analysis.
To assess the similarity between cases in a multivariate dataset, PCA starts from the characteristics (or combinations thereof) that vary most between cases. It then uses these combinations (principal components) to reduce the number of variables in the analysis. By plotting one principal component versus another, a series of two-dimensional spaces is obtained in which all observations can be positioned. The main advantage of PCA is that a continuous space is obtained where groupings are not shown as discrete units (as opposed to a hierarchical tree or k-means clustering where one case is assigned to one and only one specific cluster). We used the RC panel to calculate the principal components and the consequent vector spaces. The resulting graphs and loadings were then used to evaluate the clustering results and to position the pore water sample extract and CGS 15943 within the space obtained based on the RC.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Reference Compounds
Grouping by dendrogram.
All RC data were at first clustered using the FI and the FIS. An overview of all FI considered significant is given in Table 2 of the Supplementary Data. Table 3 in Supplementary Data lists the FIS of all RC and of their calculated grouping parameters. As a consequence, the magnitude of the induction is taken into account during calculation and the end result will depend not only on a promoter being induced or not but also on the absolute value of the induction. The latter is dependent on the concentration and the potency of the dosed compound and its possible cytotoxicity. Moreover, in the case of FI this modus operandi does not take into account the different maximum induction levels of every promoter. Figure 1 in Supplementary Data shows the result obtained based on FIS clustering; the clustering based on FI is of similar shape. From the clustering pattern some valid clusters are showing, but in general many ambiguities are observed. Cluster 1 groups include only DNA-damaging agents, and yet other DNA-damaging compounds like Ethyl methane sulphonate (EMS) and Methyl methane sulphonate (MMS), both are known DNA alkylators (as indicated by Ada induction), are not included. Cluster 2 (PCP and toxaphene) is showing, mercury and cadmium (heavy metal), and cupper and paraquat (oxidative stressors) on the other hand are separated. Moreover, the tree has a cascade shape, which on itself is indicative of poor correlation. Basically, no clear distinction between the main groups of lesions is made.
|
The dendrogram (Fig. 1) that most closely matched a priori expectations based on known mode of action was obtained by clustering the EL5 matrix (Table 2). This hierarchical tree shows two main clusters, each of which consisted of a number of subclusters, for the most part representing different functional groups. The cluster on the left hand side of the graph joins compounds that mainly induce oxidative stress and is split up into two subclusters, respectively, joining molecules that induce the oxyR operon (represented here by the KatG promoter) and compounds that primarily induce the soxRS operon (Zwf, Soi28, and nfo promoters). The main parameter allocating compounds to the soxRS operon appears to be induction of the soi28 promoter, as all soi28 inducers are allocated to this functional group (with the exception of menadione). The cluster spanning the middle and right hand side of the graph joins DNA-damaging agents and compounds grouped under general cell lesions. The DNA-damaging agents are split into two groups, one inducing mainly the Escherichia coli SOS response and one triggering the adaptive response (Ada). The cluster designated "general cell lesions" comprises three subclusters, that is, (1) compounds inducing membrane and protein damage markers, (2) molecules showing no inductions at all on this promoter set, and (3) toxicants inducing merR, as a marker for heavy metal toxicity.
k-means clustering.
Here, we will investigate the reduction in within-cluster variance and compare clustering results with those of the selected hierarchical tree. Figure 2 (Supplementary Data) shows a plot of the number of clusters versus the within-cluster sum of squares. The graph levels off at three and seven clusters, indicating a lowering of the sum of squares descent rate at these points. Therefore, this study will assess the validity of the dendrogram by setting k = 3 to show the main clusters and k = 7 to enhance the resolution within the clusters at k = 3.
|
The composition of the clusters, for both three and seven groups on the EL5 matrix, is given in Figure 1. Setting k = 3 captures the main dendrogram clusters except for the oxyR group. Probably because N,N-di-ethylformamide next to KatG also induces Zwf and Nfo, it is joined to the soxRS group (k-means cluster A). Menadione, H2O2, and N-methyl pyrrolidone (tree cluster 4) together with the cell lesion cluster outline k-means cluster B. k-means cluster C equals the tree DNA damage cluster. When k is set to 7, k-means groups the same compounds as defined in the dendrogram, except for toxaphene and PCP (tree cluster 1), which cluster together with N,N-di-ethylformamide. This can be explained by the induction of MicF and ClpB by N,N-diethylformamide, a trait it shares with toxaphene and PCP, next to the oxidative stress markers originally positioning it as oxidative stressor in the dendrogram.
Principal Component Analysis
Reference compounds.
The same data set (Table 2) was used for PCA. The PCA loading plots of the principal components explaining more than 10% of the variability in the data set are shown in Figure 2 (see also Fig. 3 and Table 4 both in Supplementary Data, respectively, depicting the scree plot, and the loadings of the first seven principal components). Note that because of (near to) identical profiles, Aflatoxin B1 and Mitomycin C coincide on the plots, as do Ascorbic acid with FeSO4 and H2O2 with N-methyl pyrrolidone. Factor 1 explains 32% of the variability of the data set; factor 2 and factor 3, respectively, explain 18 and 12%. All other factors explain less then 10% and their plots add little extra information (plots not shown). Factor 1 mainly discriminates DNA damage and general cell lesions from the oxidative stress clusters. Factor 2 adds resolution between DNA damage and the cell lesion cluster and between the different groups and molecules within the oxidative stress cluster. Additionally, this factor separates the noninducers (cluster 2) from other compounds in the cell lesion cluster. Factor 2 confirms the k-means clustering of the oxyR part of the oxidative stress group with members of the cell lesion cluster, both for k = 3 and k = 7. Factor 3 adds resolving power to the oxidative stress cluster, regrouping oxyR and separating this group from the soxRS inducing compounds. Note that the soxRS group as such is not found on the two-dimensional plots. The reason for this is best discernible in the three-dimensional plot (Figure 2D) where factor three can be seen to lift paraquat and cupper from the soxRS group up in the oxyR cluster. Both compounds indeed coinduce KatG, the distinctive marker for oxyR.
Tester samples.
Two tester samples were positioned post hoc within the principal component vector space calculated based on the RC. The location 3 C18 sample was shown to mainly induce MicF, ClpB, and UmuDC. Factor 1 causes the sample to move in the direction of the DNA-damaging (UmuDC) and cell lesion compounds (MicF and ClpB). Factor 3 shifts its position in the direction of some of the soxRS molecules and PCP and toxaphene. These are compounds coinducing the MicF and ClpB markers. As factor 2 is neutral to this sample, the factor 1 versus factor 3 plot (Fig. 2B) places this extract together with toxaphene and PCP (cluster 1 in Fig. 1). As for CGS 15943, factors 1 and 2 and the factor 1 versus factor 3 plots (Figs. 2A and 2B) indicate oxidative and cell lesion effects. The factor 2–factor 3 graph (Fig. 2C) shows the DNA-damaging potential. Exception is made for methanol; the same graph positions the molecule away from all other oxidative stressors (clusters oxyR and n° 5). It is indeed the only compound showing KatG induction (oxyR) combined with cell lesion (MicF, ClpB, and UspA) and DNA damage markers (RecA and UmuDC).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
This study used a multiendpoint bacterial gene-profiling assay in combination with a phased multivariate approach to create a classification system based on RC with known toxicological record. Two distinctively different clustering techniques were evaluated, hierarchical tree and k-means clustering, and additionally a PCA approach was used.
Quantitative gene induction assays offer a bundle of information including not only the nature or identity of the genes induced but if a dose–response approach was used, also the level of the induction in comparison to the reference system. Using all the information available, that is, the actual fold induction values of every promoter, dendrogram clustering algorithms failed to produce a toxicologically logical hierarchical tree. Hence, data were transformed into an induced–not induced table reducing the information down to the untainted mechanism of action level.
Consequently, a logical relationship was obtained, unraveling the main modes of action available through the bacterial assay. Compounds causing oxidative stress are separated from DNA-damaging agents and molecules causing other cell lesions. The latter is the fuzziest cluster, collecting membrane and protein damaging agents, as well as heavy metal like stressors and the two noninducers included in the reference set. Although many of the RC induce markers from different functional groups, clearly the dendrogram approach succeeds in disentangling important mechanistic differences. This is illustrated by the partition of the oxidative stress cluster in the oxyR and soxRS regulon subclusters, the clear distinction between the adaptive response and SOS response in case of DNA damage and the separation of toxaphene–PCP group from the remainder of the cell lesion group. In a complementary approach, the compounds are grouped using k-means clustering. The number of clusters was set to two distinct values, respectively, three and seven, based on the descending relationship between the within-cluster sum of squares and the number of clusters. k-means clustering (k = 3) identifies the original main clusters, that is, oxidative stress, DNA damage, and cell lesions, yet the oxyR group is removed from the original oxidative stress cluster and split out over the other groups. N,N-diethyl formamide is shifted to the soxRS group, thus forming the new k-means oxidative stress cluster. The remainder of oxyR is grouped within the cell lesion cluster. On one hand we believe that doing so can be toxicologically and mathematically argumented, on the other defining these compounds as oxidative stressors seems more logical. It does illustrate, however, that using only one clustering algorithm will produce a solution that could overlook mixed modes of action and thus show an oversimplified classification. Setting k equal to seven, confirms the original tree and even adds more resolution by defining the PCP–toxaphene group with N,N-diethyl formamide as a separate cluster, which is an improvement compared with the initial solution.
By consequence of the nature of both algorithms one compound can only be placed at one position of a hierarchical tree or in one k-means cluster. Although distances between clusters and compounds can be calculated and compared, one compound will only be assigned to one distinctive cluster and hence the resolution of complex modes of action spanning more then one characteristic is lost. Moreover, both algorithms make use of only one point in the dose–response curve, that is, the top exposure dose, and hence a lot of information is not taken into account. PCA can overcome part of these limitations and has an extra important advantage over traditional clustering techniques. A set of compounds can be used as a training set to calculate a fixed vector space in which afterward unknown samples/toxicants can be projected. As the calculated vector space is a continuum, PCA offers full resolution within the data set and is not hindered by compounds with mixed modes of action that cannot unambiguously be placed within one group of chemicals. Depending on the principal components plotted, the PCA analysis of the reference set confirms most of the clusters and subclusters of the initial dendrogram, albeit in clearly identifiable plotted groups or in the (uni)directional placement of individual compounds sharing common mechanisms of action. Projection of two tester samples in the PCA plane, that is, CGS19543, a compound on which only scarce toxicological information is publicly available, and an environmental sample, loc 3 C18, a C18 pore water extract from a effluent driven river basin in Belgium, illustrates the potential of the PCA and clustering approach.
CGS15943 has a mixed mode of action, showing oxidative stress, DNA damage, MicF and ClpB induction, and cell cycle disturbance (UspA). The different principal component combinations position the compound as would be expected from this complex mode of action pattern.
Also the loc 3 C18 sample is positioned as would be expected from the obtained inductions, showing DNA damage and general cell lesions. The position close to PCP and toxaphene seems the most interesting and shows the potential of this approach the best. Previous bacterial gene induction studies showed good correlation with whole organism testing at the same site (Dardenne et al., 2007b). Moreover, the sample was previously shown to cause diminished reproduction in Danio rerio (Smolders et al., 2002b, 2003) and to provoke osmolar imbalance and altered dry weight/wet weight ratios in Dreissena polymorpha (Smolders et al., 2002a, 2004) as well as impaired growth rate of Chironomus riparius larvae. Similar effects were reported in studies on PCP and toxaphene. PCP was shown to have adverse effects on reproduction of Japanese medaka (Zha et al., 2006) and is suspect of having long-term effects on ewe reproduction (Beard et al., 1999). Toxaphene studies on yellowtail flounder (Scott et al., 2002) indicate impact on liver function with possible adverse effect on reproduction, whereas Danio rerio was shown to be negatively impacted in its oviposition (FahraeusVanRee and Payne, 1997). Other authors investigated the effects of PCP and other phenolic compounds on aquatic invertebrates and showed that effects on metabolic rate are linked to the degree of narcotic toxicity. The narcotic type of toxicity and the consequent MicF induction (membrane disturbance) might be causative to the osmolar imbalance and the altered wet/dry weight ratios reported earlier by Smolders et al. in zebra mussel.
This study showed that bacterial gene induction profiles of a preselected set of RC can be grouped by principal mode of action using multivariate exploratory techniques, that is, hierarchical tree and k-means clustering. The more complex nature of some gene induction profiles, indicating mixed modes of action, can be analyzed by PCA and the resulting vector space is used to position unknown environmental samples or toxicants. This kind of mechanistic approach can be part of the solution to the vast toxicological task required for REACH compliance (e.g., chemical grouping and read-across) and to the ever increasing need for reliable and relevant in vitro testing systems replacing or complementing whole organism testing in (eco)toxicological assessments.
|
SUPPLEMENTARY DATA |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Supplementary data are available online at http://toxsci.oxfordjournals.org/.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Adams SM, Hill WR, Peterson MJ, Ryon MG, Smith JG, Stewart AJ. Assessing recovery in a stream ecosystem: Applying multiple chemical and biological endpoints. Ecol. Appl. (2002) 12:1510–1527.[CrossRef]
Beard AP, Bartlewski PM, Rawlings NC. Endocrine and reproductive function in ewes exposed to the organochlorine pesticides lindane or pentachlorophenol. J. Toxicol. Environ. Health A (1999) 56:23–46.[CrossRef][Web of Science][Medline]
Bervoets L, Blust R. Metal concentrations in water, sediment and gudgeon (Gobio gobio) from a pollution gradient: Relationship with fish condition factor. Environ. Pollut. (2003) 126:9–19.[CrossRef][Medline]
Combes R, Barratt M, Balls M. An overall strategy for the testing of chemicals for human hazard and risk assessment under the EU REACH system. Altern. Lab. Anim. (2003) 31:7–19.[Medline]
Dardenne F, Nobels I, De Coen W, Blust R. Dose–response relationships and statistical performance of a battery of bacterial gene profiling assays. Appl. Microbiol. Biotechnol. (2007a) 75:223–234.[CrossRef][Web of Science][Medline]
Dardenne F, Smolders R, DeCoen W, Blust R. Prokaryotic gene profiling assays to detect sediment toxicity: Evaluating the ecotoxicological relevance of a cell-based assay. Environ. Sci. Technol. (2007b) 41:1790–1796.[Medline]
Delihas N, Forst S. MicF: An antisense RNA gene involved in response of Escherichia coli to global stress factors. J. Mol. Biol. (2001) 313:1–12.[CrossRef][Web of Science][Medline]
Diez A, Gustavsson N, Nystrom T. The universal stress protein a of Escherichia coli is required for the resistance to DNA damaging agents and is regulated by a RecA/FtsK dependent regulatory pathway. Mol. Microbiol. (2000) 36:1494–1503.[CrossRef][Web of Science][Medline]
Eggen RIL, Segner H. The potential of mechanism-based bioanalytical tools in ecotoxicological exposure and effect assessment. Anal. Bioanal. Chem. (2003) 377:386–396.[CrossRef][Web of Science][Medline]
Escher BI, Hermens JLM. Modes of action in ecotoxicology: Their role in body burdens, species sensitivity, QSARs, and mixture effects. Environ. Sci. Technol. (2002) 36:4201–4217.[Medline]
FahraeusVanRee GE, Payne JF. Effect of toxaphene on reproduction of fish. Chemosphere (1997) 34:855–867.[CrossRef][Web of Science][Medline]
Fawcett WP, Wolf RE. Genetic definition of the Escherichia-coli zwf soxbox, the DNA-binding site for soxs-mediated induction of glucose-6-phosphate-dehydrogenase in response to superoxide. J. Bacteriol. (1995) 177:1742–1750.
Galhardo RS, Almeida CEB, Leitao AC, Cabral-Neto JB. Repair of DNA lesions induced by hydrogen peroxide in the presence of iron chelators in Escherichia coli: Participation of endonuclease IV and Fpg. J. Bacteriol. (2000) 182:1964–1968.
Gineinah MM, Nasr MN, Abdelal AM, El Emam AA, Said SA. Synthesis and antiinflammatory evaluation of new 2-and 3-substituted 1,2,4-triazolo[4,3-c] and [1,5-c]quinazoline derivatives. Med. Chem. Res. (2000) 10:243–252.[Web of Science]
Harms H, Wells MC, van der Meer JR. Whole-cell living biosensors—Are they ready for environmental application? Appl. Microbiol. Biotechnol. (2006) 70:273–280.[CrossRef][Web of Science][Medline]
Hofer T, Gerner I, Gundert-Remy U, Liebsch M, Schulte A, Spielmann H, Vogel R, Wettig K. Animal testing and alternative approaches for the human health risk assessment under the proposed new European chemicals regulation. Arch. Toxicol. (2004) 78:549–564.[CrossRef][Web of Science][Medline]
Houtman CJ, Cenijn PH, Hamers T, Lamoree MH, Legler J, Murk AJ, Brouwer A. Toxicological profiling of sediments using in vitro bioassays, with emphasis on endocrine disruption. Environ. Toxicol. Chem. (2004) 2:32–40.
Huisman O, Dari R, Gottesman S. Cell-division control in Escherichia-coli—Specific induction of the Sos function Sfia protein is sufficient to block septation. Proc. Natl. Acad. Sci. U.S.A. (1984) 81:4490–4494.
Jantova S, Stankovsky S, Spirkova K. In vitro antibacterial activity of ten series of substituted quinazolines. Biologia (2004) 59:741–752.[Web of Science]
Johnson BT, Long ER. Rapid toxicity assessment of sediments from estuarine ecosystems: A new tandem in vitro testing approach. Environ. Toxicol. Chem. (1998) 17:1099–1106.[CrossRef][Web of Science]
Kenyon CJ, Brent R, Ptashne M, Walker GC. Regulation of damage-inducible genes in Escherichia-coli. J. Mol. Biol. (1982) 160:445–457.[CrossRef][Web of Science][Medline]
Kim G, Oh TJ. 2-Mercaptoethylamine, radioprotector, inhibits the induction of the oxidative stress inducible (soi) gene by paraquat in Escherichia coli. Pharm. Res. (2000) 42:429–433.[CrossRef][Web of Science]
Kitagawa M, Wada C, Yoshioka S, Yura T. Expression of Clpb, an analog of the Atp-dependent protease regulatory subunit in Escherichia-coli, is controlled by a heat-shock sigma-factor (SIGMA-32). J. Bacteriol. (1991) 173:4247–4253.
Kohler S, Belkin S, Schmid RD. Reporter gene bioassays in environmental analysis. Fresenius J. Anal. Chem. (2000) 366:769–779.[CrossRef][Medline]
Landini P, Hajec LI, Volkert MR. Structure and transcriptional regulation of the Escherichia coli adaptive response gene aidB. J. Bacteriol. (1994) 176:6583–6589.
Livrelli V, Lee IW, Summers AO. In vivo DNA-protein interactions at the divergent mercury resistance (mer) promoters. 1. Metalloregulatory protein merr mutants. J. Biol. Chem. (1993) 268:2623–2631.
Pfeiffer WD, Hetzheim A, Pazdera P, Bodtke A, Mucke J. Synthesis and reactivity of 1,2,4-triazolo[1,5-c]quinazolines. J. Heterocycl. Chem. (1999) 36:1327–1336.[CrossRef][Web of Science]
Reifferscheid G, Hell J. Validation of the SOS/umu test using test results of 486 chemicals and comparison with the Ames test and carcinogenicity data. Mutat. Res. Genet. Toxicol. (1996) 369:129–145.
Sambrook J, Fritsch EF, Maniatis T. Molecular Cloning, A Laboratory Manual (1989) New York: Cold Spring Harbor Laboratory Press.
Scott KD, Fahraeus-Van Ree GE, Parrish CC. Sex differences in hepatic lipids of toxaphene-exposed juvenile yellowtail flounder (Pleuronectes ferrugineus storer). Ecotoxicol. Environ. Saf. (2002) 51:168–176.[CrossRef][Web of Science][Medline]
Smith BT, Walker GC. Mutagenesis and more: umuDC and the Escherichia coli SOS response. Genetics (1998) 148:1599–1610.
Smolders R, Bervoets L, Blust R. Transplanted zebra mussels (Dreissena polymorpha) as active biomonitors in an effluent-dominated river. Environ. Toxicol. Chem. (2002a) 21:1889–1896.[CrossRef][Web of Science][Medline]
Smolders R, Bervoets L, Blust R. In situ and laboratory bioassays to evaluate the impact of effluent discharges on receiving aquatic ecosystems. Environ. Pollut. (2004) 132:231–243.[CrossRef][Medline]
Smolders R, Bervoets L, De Boeck G, Blust R. Integrated condition indices as a measure of whole effluent toxicity in zebrafish (Danio rerio). Environ. Toxicol. Chem. (2002b) 21:87–93.[CrossRef][Web of Science][Medline]
Smolders R, De Boeck G, Blust R. Changes in cellular energy budget as a measure of whole effluent toxicity in zebrafish (Danio rerio). Environ. Toxicol. Chem. (2003) 22:890–899.[CrossRef][Web of Science][Medline]
Sorensen SJ, Burmolle M, Hansen LH. Making bio-sense of toxicity: New developments in whole-cell biosensors. Curr. Opin. Biotechnol. (2006) 17:11–16.[CrossRef][Web of Science][Medline]
Tartaglia LA, Storz G, Ames BN. Identification and molecular analysis of Oxyr-regulated promoters important for the bacterial adaptation to oxidative stress. J. Mol. Biol. (1989) 210:709–719.[CrossRef][Web of Science][Medline]
Waring JF, Ulrich RG. The impact of genomics-based technologies on drug safety evaluation. Annu. Rev. Pharmacol. Toxicol. (2000) 40:335–352.[CrossRef][Web of Science][Medline]
White PA, Blaise C, Rasmussen JB. Detection of genotoxic substances in bivalve molluscs from the Saguenay Fjord (Canada), using the SOS chromotest. Mutat. Res. Genet. Toxicol. Environ. Mutagen. (1997) 392:277–300.[CrossRef]
White PA, Rasmussen JB, Blaise C. A semi-automated, microplate version of the SOS Chromotest for the analysis of complex environmental extracts. Mutat. Res. Environ. Mutagen. Relat. Subj. (1996) 360:51–74.[CrossRef]
Witkin EM. RecA protein in the SOS response—Milestones and mysteries. Biochimie (1991) 73:133–141.[Medline]
Yim HH, Brems RL, Villarejo M. Molecular characterization of the promoter of OsmY, an RPOS dependent gene. J. Bacteriol. (1994) 176:100–107.
Zha JM, Wang ZJ, Schlenk D. Effects of pentachlorophenol on the reproduction of Japanese medaka (Oryzias latipes). Chem. Biol. Interact. (2006) 161:26–36.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||