Genomic profiles and predictive biological networks in oxidant-induced atherogenesis
C. D. Johnson1,
Y. Balagurunathan1,
K. P. Lu1,
M. Tadesse2,
M. H. Falahatpisheh1,
R. J. Carroll2,
E. R. Dougherty1,
C. A. Afshari3 and
K. S. Ramos1
1 Center for Environmental and Rural Health, Texas A&M University, College Station, Texas 77843
2 Department of Statistics, Texas A&M University, College Station, Texas 77843
3 National Institute of Environmental Health Sciences, Research Triangle, North Carolina 27709
 |
ABSTRACT
|
---|
Atherogenic stimuli trigger complex responses in vascular smooth muscle cells (VSMCs) that culminate in activation/repression of overlapping signal transduction cascades involving oxidative stress. In the case of benzo[a]pyrene (BaP), a polycyclic aromatic hydrocarbon present in tobacco smoke, the atherogenic response involves interference with redox homeostasis by oxidative intermediates of BaP metabolism. The present studies were conducted to define genomic profiles and predictive gene biological networks associated with the atherogenic response of murine (aortic) VSMCs to BaP. A combined oxidant-antioxidant treatment regimen was used to identify redox-sensitive targets during the early course of the atherogenic response. Gene expression profiles were defined using cDNA microarrays coupled to analysis of variance and several clustering methodologies. A predictor algorithm was then applied to gain insight into critical gene-gene interactions during atherogenesis. Supervised and nonsupervised analyses identified clones highly regulated by BaP, unaffected by antioxidant, and neutralized by combined chemical treatments. Lymphocyte antigen-6 complex, histocompatibility class I component factors, secreted phosphoprotein, and several interferon-inducible proteins were identified as novel redox-regulated targets of BaP. Predictor analysis confirmed these relationships and identified immune-related genes as critical molecular targets of BaP. Redox-dependent patterns of gene deregulation indicate that oxidative stress plays a prominent role during the early stages of BaP-induced atherogenesis.
analysis of variance; atherosclerosis; bioinformatics; cDNA microarray; polycyclic aromatic hydrocarbons; vascular smooth muscle cells
 |
INTRODUCTION
|
---|
ATHEROSCLEROSIS IS THE LEADING cause of morbidity and mortality in industrialized nations. The formation of atherosclerotic lesions involves migration of vascular smooth muscle cells (VSMCs) from the tunica media toward the vessel lumen and uncontrolled cellular proliferation (37, 41). Oxidants and pro-oxidants present in tobacco smoke increase the formation of atherosclerotic lesions in laboratory animals and humans (17, 40). To date, however, the molecular mechanisms of vascular gene deregulation following environmental injury remain unclear.
Using benzo[a]pyrene (BaP) as a model atherogen, this laboratory has shown that transcriptional interference of selected genes is a critical event in the induction of proliferative (i.e., atherogenic) phenotypes in rat, mouse, quail, and human VSMCs (40). Vascular cytochrome P-450 enzymes metabolize BaP to arene oxides that spontaneously oxidize to phenols or dihydrodiols (30). BaP and its metabolites trigger a complex cellular response that culminates in activation of overlapping signal transduction cascades involving oxidative stress and aryl hydrocarbon receptor (Ahr) (4, 21). The atherogenic response to BaP shares striking homologies to the carcinogenic process during which BaP and its metabolites induce oxidative stress and DNA damage (14, 29, 32). Thus common pathogenetic links may exist between atherogenesis and carcinogenesis.
The interactive gene networks responsible for induction of atherogenic VSMC phenotypes following oxidative environmental injury have not yet been identified. To understand the complex nature of the atherogenic process, it is critical to use a holistic approach that examines global patterns of gene expression in a contextual manner. Although there is nearly limitless variance in patterns of global gene expression at any given point in time, VSMCs exhibit a finite number of phenotypic states that either support vascular wall homeostasis or give rise to pathogenetic changes. Differential display and serial analysis of gene expression have been used to screen large numbers of cDNA clones (27, 28). Several limitations render these techniques nonconducive to large-scale expression analysis, and therefore complementary nucleotide hybridization technology is now used to monitor the expression of thousands of unique sequences at a given point in time. Although mRNA is not the final product of a gene, transcription is a critical component in the regulation of protein expression and therefore provides an ideal point of investigation.
The present studies were conducted to define genomic profiles and predictive networks of biological activity during the early stages of the atherogenic response to BaP. Murine (aortic) VSMCs were challenged for 24 h with BaP alone, or in the presence of N-acetylcysteine (NAC) and allowed to recover from chemical treatments for 1 wk. This exposure regimen does not reflect all forms of oxidative stress encountered by the arterial wall but is a reproducible stress that mimics the physiological oxidative stresses induced by tobacco smoke and other forms of environmental injury. Evidence is presented that deregulation of VSMC gene expression during the early phases of the adaptive response to oxidative injury involves disruption of redox-regulated gene expression. Of relevance was the identification of fyn proto-oncogene, ndr3, and hemochromatosis gene as predictors of Ahr, and the emergence of histocompatibility complex associated genes as predictors of cyclin D and lysyl oxidase.
 |
Materials And Methods
|
---|
Cell Culture and Chemical Treatments
Cells were isolated from the thoracic aorta of three C57BL/6J (6 wk old) naive mice and maintained in serial culture using standard methodologies (38). Aorta was used because this vessel is susceptible to atherosclerotic lesion formation in mice (10) and is the largest vessel from which sufficient numbers of cells can be isolated with relative ease. In the first set of studies, G0 synchronized cultures at passage 12 (75% confluence) were released into growth by addition of fetal bovine serum (10%) in the presence of BaP (3 µM; Sigma-Aldrich) or dimethyl sulfoxide (DMSO, 0.0075%; Sigma-Aldrich) for 24 h. This regimen has been associated with induction of atherogenic phenotypes in the absence of excess rates of apoptotic cell death (21, 34). A separate set of cultures was pretreated for 1 h with 0.5 mM NAC (Sigma-Aldrich), a water-soluble antioxidant and precursor of cellular glutathione, dissolved in culture medium prior to BaP treatment to increase glutathione levels and enhance antioxidant activity (20). Cultures were allowed to recover for 1 wk after chemical treatments to identify genes involved in the early adaptive response to atherogenic insult. At the end of the 7-day recovery period, plates from each treatment group along with DMSO control were processed for isolation of 500 or 1,500 µg of total RNA, respectively. Total RNA was isolated from cells using TRI-Reagent (Molecular Research Center, OH) according to standard protocols. Isolation of poly(A)+ mRNA from total RNA was completed using Oligotex mRNA midi Kits (Qiagen, Valencia, CA). The quality of the RNA was determined by gel electrophoresis and Bioanalyzer methodology (Agilent, Palo Alto, CA).
In the second set of studies, G0 synchronized cultures at passage 12 (75% confluence) were released into growth by addition of fetal bovine serum (10%) in the presence of 3 µM BaP (Sigma-Aldrich), BaP 7,8-diol (Midwest Research Institute), BaP 3,6-quinone (Midwest Research Institute), or DMSO (Sigma-Aldrich) for 24 h. A separate set of cultures was pretreated for 1 h with 0.5 mM NAC before hydrocarbon challenge. At the end of a 1-wk recovery period, plates from each treatment group along with DMSO controls were processed for isolation of total RNA using TRI Reagent (Molecular Research Center).
Microarray Hybridizations
Mouse cDNA arrays developed in-house at National Institute of Environmental Health Sciences (NIEHS) according to the specifications of DeRisi et al. (6) were used for gene expression profiling. A complete listing of the 8,976 transcripts on this chip is available at http://dir.niehs.nih.gov/microarray/chips.htm. The spotted cDNAs were derived from a collection of sequence verified IMAGE clones that spanned the 5' end of the gene and ranged in size from 500 to 2,000 bp (Incyte Genomics, Palo Alto, CA). M13 primers were used to amplify insert cDNAs from purified plasmid DNA in a 100-µl PCR reaction mixture. An aliquot of the PCR products (10 µl) was separated on 2% agarose gels to ensure quality of the amplification. The remaining product was purified by ethanol precipitation, resuspended in ArrayIt buffer (Telechem) and spotted onto poly-L-lysine-coated glass slides using a modified, robotic DNA arrayer (Beecher Instruments). RNA samples representing three treatment groups and one set of controls were processed at the NIEHS Microarray Center. Each comparison between three treatment groups and one control was duplicated four times for 12 independent hybridizations. Each poly(A)+ RNA sample (24 µg) was labeled with cyanine-3 (Cy3) or cyanine-5 (Cy5)-conjugated dUTP (Amersham) by reverse transcription using SuperScript (Invitrogen) and oligo-dT (Amersham). The fluorescently labeled cDNAs were mixed and hybridized simultaneously to the cDNA microarray chip. Each treatment/control pair was hybridized to four arrays employing fluor reversal accomplished by labeling the control sample with Cy3 in two hybridizations and with Cy5 in the other two hybridizations. The cDNA chips were scanned with an Axon 4000b scanner (Axon Instruments) using independent laser excitation of the two fluors at 532- and 635-nm wavelengths for the Cy3 and Cy5 labels, respectively.
The raw pixel intensity images were analyzed using the ArraySuite v1.3 extensions of the IPLab image processing software package (Scanalytics). These extension routines were developed at National Human Genome Research Institute and the mathematical and image-processing methods used described in Chen et al. (2, 3). These methods locate target sites on the array and determine probe intensity. This program uses methods previously described by Chen et al. (2) to locate target sites on the array and determine probe intensity. The mouse arrays contained 8,976 cDNAs corresponding to 8,976 unique genes. The ratio intensities for all spots were used to fit a probability distribution to the ratio intensity values and to estimate the normalization constants (m and c) that this distribution provides. The constant m, which provides a measure of the intensity gain between the two channels, ranged from 0.78 to 1.28 for all arrays, indicating that the channels were balanced near one. For each array, the ratio intensity values were normalized to account for the imbalance between the two fluorescent dyes by multiplying the ratio intensity value by m. The other constant, c, estimates the coefficient of variation for the intensity values of the two samples. All arrays in this analysis had a c value of 0.2 or less. The probability distribution that is fit to the data was used to calculate a 99% confidence interval for the ratio intensity values. It has been previously determined that significant autofluorescence of the gene features on the array, attributed to spotting solution, occurs at high scanning power. To correct for this problem, the pixel intensity level of "blank" spots comprised of spotting solution was measured. The data were then filtered to provide a cutoff at the intensity level just above the blank measurement values to remove from further analyses those genes having one or more intensity values in the background range. All ratio values were transformed to a log-base 2 scale; yielding an approximately normally distributed dataset and allowing treatment of inductions or repressions of identical magnitude as numerically equal but with opposite sign. The gene expression profiles for each treatment were measured using microarray and a subset of genes on the array confirmed using real-time PCR.
Analysis of Variance
The assumption of normality on the log scale was inspected using standard graphical and statistical methods. An ANOVA model for treatment effect was fitted on each of the 8,976 transcripts examined to determine differentially expressed genes across treatments (11). All possible pairwise contrasts between treatments were also examined.
Clustering
Clustering techniques can be divided into two general classes: supervised and unsupervised. In supervised clustering, existing biological information was used to guide the classification of genes affected by chemical treatment. The unsupervised methods (i.e., data driven) considered here included k-means, fuzzy c-means, and hierarchical clustering. The k-means clustering is an iterative procedure which partitions the data into k groups, such that the observations within a cluster are more similar than those in different clusters (7). The number of clusters is defined, based on the ratio of the between- and within-cluster sum-of-squares for various values of k. The major drawbacks of k-means clustering are that the number of clusters is fixed in advance with no knowledge of how many clusters exist, as well as its dependence on cluster center point groups prone to unstable groupings. To correct for this, the algorithm was run several times with different starting points, and the partition obtained a high proportion of the time selected. Fuzzy c-means is considered an extension of k-means clustering which assumes that boundaries between subgroups are not well defined and therefore do not force each data point to belong to one and only one cluster. Instead, each data point is assigned to a cluster with some degree of membership. Hierarchical clustering, on the other hand, creates a dendrogram where the branch lengths represent the degree of similarity between gene sets. This approach can be agglomerative or divisive, where each gene is treated as a cluster and merges at each step to similar sets, to create a single large cluster, or a large cluster splits dissimilar sets into smaller clusters, respectively. There are several ways of merging or splitting the clusters based on the distance between members, with commonly used metrics including single, complete, and average linkage. A major shortcoming of hierarchical clustering is the deterministic nature of the analysis with no opportunity for reallocation of genes that might have been grouped "incorrectly" at an earlier stage. Another potential problem is that as clusters grow in size, the expression vector representing a cluster might no longer represent any of the genes within the cluster.
Real-time PCR
The double-stranded DNA binding dye method was used to measure mRNA levels of selected genes. Primers were chosen using Oligo 4.0 (National Biosciences, Plymouth, MN) and designed to amplify all major variants of the gene. Primers were as follows (5'-3'): lymphocyte antigen 6c, forward GCGAAGACCTCTGCAATGC, reverse AGCTCAGGCTGAACAGAAGCA; phosphoprotein 1, forward AGGAAACCAGCCAAGGACTAACT, reverse GCAATGCCAAACAGGCAAA; histocompatibility class I-Q region loc. 7, forward TCAGGCTCCCTCCGGAAACT, reverse AGTATTGGGAGCGGGAGACA; histocompatibility class I-L region, forward CCTCCTCCGTCCACTGACTCT, reverse TCCAATGATGCCATAGCTCAAG; and histocompatibility class I-K region, forward CCAGGTAGGCCCTGAGTCTCT, reverse GGCGCTGATCAACCAAACAC. Reverse transcription was carried out in a final volume of 20 µl containing 1x RT-PCR buffer, 1 mM of each deoxynucleotides triphosphate, 5 mM MgCl2, 1 U/µl RNase inhibitor, 2.5 U/ml reverse transcriptase, 50 U oligo d(T), and 200 µg of total RNA. Samples were incubated at room temperature for 10 min, then 42°C for 30 min, followed by heat inactivation at 99°C for 5 min and cooling at 5°C for 5 min. Real-time PCR amplification was performed using ABI Prism 7700 Sequence Detection System (Perkin-Elmer Applied Biosystems). For each run, 25 ml of 2x SYBR Master mix (Perkin-Elmer Applied Biosystems) and 0.4 µM of both forward and reverse primer along with 10 µl of each appropriate transcriptase samples were mixed. The thermal cycling conditions comprised an initial denaturation step at 95°C for 10 min, 50 cycles at 95°C for 15 s, and 65°C for 1 min.
Predictor Calculations
The procedure proposed by Kim et al. (22, 23) was used for prediction analysis. A subset of
200 genes with the smallest main effect ANOVA P values was selected, and their profiles were used to predict the behavior of target genes specified based on expected biological relevance (16). The target genes selected were chosen on the basis of known biological relevance in VSMCs, and included antioxidant protein, apoptosis inhibitor, Ahr, cyclin D, glutathione peroxidase, rat SMC growth factor-responsive gene, histocompatibility class I-L region, insulin-like growth factor binding protein 5 (IGFBP-5), lymphocyte antigen 6 complex, lysyl oxidase, MAP kinase-activated protein kinase 2, matrix metalloproteinase-2, thioredoxin, and thioredoxin peroxidase. The goal was to identify small sets of genes whose transcriptional states are predictive of a given target. The predictor genes may lie upstream or downstream from a target within the gene interaction network, or the relationship to the target may be based on chains of interaction among various intermediates (23). There was no assumption of causality in the prediction method, and its sole focus was to identify sets of genes that may be associated with the target gene.
The algorithm started by categorizing the transcript levels into ternary expression data: -1 for downregulated, 0 for invariant, and +1 for upregulated genes. The data were then divided into a training set and a test set. Based on the training data, the conditional probability that the target gene takes on one of the three transcriptional states was calculated for all possible patterns of the predictor genes and the predicted target value defined as the state with the largest conditional probability. In considering a predictor set with two genes, the results can be summarized as follows
where t1, ... , t9 equal -1, 0, or +1. The analysis then reverted to the test data to examine the performance of these predictors. The error for each of the predictor functions is given by |Tobs - T*|, where Tobs is the observed and T* is the predicted transcriptional state, which could be the optimal predictor state, T
, obtained by the designed filter or the reference predictor state, Tµ, obtained by the reference filter.
The above procedure was repeated numerous times by randomly splitting the data into training and test sets, in a fixed proportion. The test error was estimated by averaging the prediction error across all iterations, and this error was computed for all possible predictor combinations. The performance of a set of predictors was determined by a statistic known as the coefficient of determination (COD) (8, 23). This coefficient measured the degree to which the transcriptional levels of a set of genes can be used to improve the prediction of the transcriptional state of a target gene relative to the best possible prediction in the absence of predictors. In this case, we used mean of the target gene as the reference metric, (its transcriptional state represented by Tµ). The COD (
) is defined by the following relation
where
· is the average error for the best predictor in absence of observation and 
is average error due to the optimal predictor, designed. The errors with respect to n observations are given by
The higher the COD
(close to 1), the more accurate would be the prediction of the targets transcriptional state, i.e., the higher the degree of relationship between the target and predictor genes. All possible combinations of 1, 2, and 3 gene predictors for the chosen targets were studied with the number of possible predictors runs in the order of millions for 1, 2, and 3 gene combinations, respectively, for each of the targets. Predictors were ordered with respect to their errors and the COD values, and the analysis was focused on COD values greater than 0.9 and a test error less than 0.05. Information obtained is suggestive of biological commonality between predictor genes and the specified target gene.
 |
Results
|
---|
Genomic Profiling
Statistical analysis.
To define genomic profiles during the early phase of the atherogenic response, murine VSMCs were challenged for 24 h with 3 µM BaP alone, or in the presence of 0.5 mM NAC, and allowed to recover under normal culture conditions for 1 wk before mRNA isolation. Treatment-related changes in VSMC gene expression are depicted as a Venn diagram in Fig. 1 and itemized in the Supplemental APPENDIX 1. (APPENDIX 1 and APPENDIX 2 are available as a data supplement, published online at the Physiological Genomics web site.1
) The transcription levels of 819 clones were changed significantly in BaP-treated cells, with 673 of these clones represented exclusively within this group. Affected genes included those involved in apoptosis (DNase inhibited and death-associated kinase), growth regulation [platelet-derived growth factor receptor-
(PDGFr-
) and IGF], and immune modulation (lymphocyte antigen 6 complex, IIGP protein and Mx2). These genes behaved as redox-regulated targets in that they were altered by pro-oxidant treatment and neutralized by the antioxidant.

View larger version (33K):
[in this window]
[in a new window]
|
Fig. 1. Venn diagram of genes significantly expressed in redox-activated vascular smooth muscle cells (VSMCs). Treatment-related changes relative to DMSO were identified using Students t-test. Red is representative of genes significantly expressed in BaP-treated cells, yellow indicates of genes significantly expressed in BaP and BaP/NAC groups, green indicates genes expressed only in the BaP/NAC group, light blue indicates genes expressed in cells treated with NAC and BaP/NAC, blue indicates genes expressed only in the NAC group, purple indicates genes expressed in BaP- and NAC-treated cells, and white indicates genes expressed under all treatment conditions. NAC, N-acetylcysteine; BaP, benzo[a]pyrene.
|
|
NAC modulated a total 336 clone transcriptional states, with 255 genes exclusively modulated by antioxidant treatment. These genes included procollagen type III and X, apoptosis inhibitor, mannosidase 1
, and ras GTPase activating protein. Genes within the NAC group also exhibited reciprocal redox control. Of 60 clones altered by BaP or NAC, most affected clones were expressed sequence tags (ESTs) preferentially downregulated by redox stress. Combined BaP/NAC treatment altered the expression of 395 clones relative to DMSO control. Of these, 283 clones were representative of a subset of ESTs strictly regulated by the extreme redox microenvironment induced by combined chemical treatments. Genes in this group included MSSP, a c-Myc binding protein, Y box protein, cyclin D, superoxide dismutase, transformation related protein 5, Ndr1 related protein, and tumor necrosis factor receptor 1. Sixty-five clones were represented in both the BaP and BaP/NAC groups, including death-associated kinase 2, interferon-induced guanylate binding protein GBP-2, feminization 1 homolog, growth factor receptor bound protein, histocompatibility class I-K, -L, and -Q region locus, and HMG-I(Y). These genes were unaffected by antioxidant treatment either due to lack of redox sensitivity, or deficient antioxidant activity. A subset of 28 genes was identified that changed exclusively in the NAC and NAC/BaP treatment groups but which were responsive to the antioxidant. The last group of genes identified included ferritin heavy chain and secreted phosphoprotein, genes highly responsive to changes in redox status.
Cluster analysis.
The k-means, fuzzy c-means, and hierarchical clustering were applied to the entire dataset to identify genes modulated by BaP in a redox-sensitive manner. Of interest was the finding that many of the genes identified as redox-regulated targets in Fig. 1 clustered together using clustering methodology (Fig. 2). The three nonsupervised methods resolved similar groups of genes and identified clones that were highly upregulated by BaP alone, unaffected by NAC, and neutralized by combined BaP/NAC treatment.
Hierarchical clustering was chosen to further resolve clusters of redox-regulated genes because the other methods forced groupings into clusters of equal size without regard to biological outcome and ignored antioxidant modification of gene expression across treatments. Euclidean distance was chosen for the similarity measure coupled to complete linkage analysis. ANOVA was employed to identify significant changes (P
0.05) in transcriptome profiles across the various treatment groups. As shown in Fig. 2, 90 cDNAs of known function were differentially expressed among all treatment groups. Of these, 16 were regulated by BaP alone, 1 by NAC, 7 by BaP or NAC, 30 by BaP and neutralized by NAC, 2 by all oxidant/antioxidant treatments, 12 by the combination of BaP and NAC, and 21 by NAC and neutralized by BaP. A dendrogram along with a colored matrix of gene expression values and profiles for the redox-regulated genes is displayed in Fig. 3. Genes with no change relative to control (ratios of 1) were colored black, upregulated gene profiles are in red (ratios greater than 1), and downregulated gene profiles are in green (ratios less than 1). The gradation in colors reflects the degree of regulation in either direction. Of interest within the context of the atherogenic response to BaP were genes that exhibited elements of reciprocal redox regulation; that is, modulation by BaP and neutralization by NAC, or vice versa. Using this approach, several interesting targets were identified, including genes involved in the immune-mediated inflammatory response (lymphocyte 6 antigen, major histocompatibility class I complex, secreted phosphoprotein, oligoadenylate synthase-like protein, and interferon-inducible genes) (Table 1). In addition, genes involved in cellular metabolism (2,4-dienoyl-coenzyme A reductase 2, carbonic anhydrase-like sequence, UDP glucose dehydrogenase, mannosidase 1
and 2
) and growth and differentiation [guanylate binding protein, IGF binding protein, EST moderately similar to actin-like protein 14D, HMG-I(Y), transcription termination factor, and zinc finger protein 14] were identified. Other interesting targets included ubiquitin conjugating enzyme, amyloid ß precursor protein, and procollagen type III-
. A graphical representation of the average log-ratio values of significantly altered genes within each cluster is shown in Fig. 4. This analysis readily identified genes that were modified by BaP and neutralized in the presence of antioxidant.

View larger version (52K):
[in this window]
[in a new window]
|
Fig. 3. Dendrogram depicting hierarchical clustering relationships for putative redox-regulated targets identified by ANOVA. A colored matrix of expression patterns for genes modified by BaP, unaffected by NAC, and differentially modified by combined BaP/NAC treatments. Amplified profiles for selected redox-regulated targets are also shown.
|
|

View larger version (23K):
[in this window]
[in a new window]
|
Fig. 4. Patterns of redox-regulated gene expression in VSMCs. A: three-dimensional representation of ANOVA selected redox-regulated targets as classified using hierarchical methodology. B: graphical representation of antioxidant-induced deviations in expression profile as defined by linear regression. C: mean average values for each hierarchical cluster identified by ANOVA as redox-regulated targets.
|
|
Real-time PCR.
To evaluate the redox sensitivity and specificity of the genomic response to BaP, and the role of oxidative intermediates of BaP metabolism in modulation of transcriptome profiles, VSMCs were challenged with equimolar concentrations of BaP, BaP 3,6-quinone, and BaP 7,8-diol. These intermediates are formed in VSMCs following cytochrome P-450-catalyzed oxidation of the parent compound (30, 31). Redox-regulated patterns of expression for lymphocyte antigen 6 complex, secreted phosphoprotein 1, histocompatibility class I-Q region locus 7, histocompatibility class I-L region, and histocompatibility class I-K region were quantified by real-time PCR (Fig. 5). These genes were chosen for further analysis because they were among the most novel subset of BaP-regulated genes in VSMCs. The same experimental design used for global gene expression analysis was used in these experiments. A 2- to 10-fold increase in the expression of all genes was observed in cells treated with 3 µM BaP for 24 h and allowed to recover for 1 wk. As expected, 0.5 mM NAC treatment significantly neutralized the induction of all target genes. In the case of lymphocyte antigen 6 complex, BaP 3,6-quinone and BaP 7,8-diol increased gene expression in a redox-dependent manner. However, BaP metabolites did not regulate the expression of other targets, suggesting that significant differences in redox sensitivity influence the expression of selected targets.

View larger version (28K):
[in this window]
[in a new window]
|
Fig. 5. Gene expression patterns of selected target genes in VSMCs treated with BaP and related oxidative metabolic intermediates. VSMCs were treated with 3 µM BaP, BaP 7,8-diol, BaP 3,6-quinone, or DMSO for 24 h. A separate set of cultures was pretreated for 1 h with 0.5 mM NAC. mRNA levels were quantified using semiquantitative RT-PCR as described in MATERIALS AND METHODS.
|
|
Computational predictor analyses.
Several clones were selected based on ANOVA P values
0.01 for further analysis using a computational target clone-predictor method. This strategy selected for genes within the dataset with high probability to behave as superior singleton predictors. This level of stringency identified 200 representative clones, a size manageable within the computational constraints of the desktop computer used (Pentium 4, 2.4 GHz, 1,024 Mb RAM). Target clones were chosen a priori based on their biological significance within the context of the atherogenic response. Three clone predictor combinations were ranked based on COD and prediction error, with those of COD greater than 0.9 and errors less than 0.05 expressed as a percentage of the time the gene was identified as a predictor for a given target (Table 2). A large number of three-clone combinations met these criteria for most targets, with one or two clones identified as predominant predictors within the sample pool. A complete listing of target-predictor clones as presented in Supplemental APPENDIX 2. As expected, target genes known to be promiscuously regulated showed the larger number of predictor sets.
Antioxidant protein, a thioredoxin-dependent peroxidase, was best predicted by lysyl oxidase (selected as a predictor 67% of the time), 3-hydroxy-3-methylglutaryl-coenzyme A synthase (67%), type VI collagen-
(33%), myristoylated alanine rich protein kinase C substrate (33%), guanidinoacetate methyltransferase (33%), claudin-7 (33%), and monocyte chemotactic protein-5 (33%). The best predictors for apoptosis inhibitor gene were thioredoxin peroxidase (100%), DNase inhibitor (67%), and interferon-inducible protein (33%). Of interest within the context of BaP atherogenesis was the identification of fyn proto-oncogene (75%), N-myc downstream regulated 3 (50%), hemochromatosis (50%), and CDC7-related kinase (50%) as major predictors of Ahr gene expression. Ahr is a basic helix-loop-helix transcription factor strongly implicated in the atherogenic response to BaP and related aromatic hydrocarbons (20, 21). Interestingly, only the voltage-dependent
-subunit of calcium channel was identified as a relatively frequent predictor of the glutathione peroxidase. For rSMC growth factor responsive protein, laminin (15%), peroxisome biogenesis factor 1 (15%), and B-cell leukemia/lymphoma 3 (15%) were identified as best predictors. Lysyl oxidase was best predicted by histocompatibility class I-Q region locus (25%), mitogen activated protein kinase 9 (21%), Ig-
3 chain C region (17%), and lysyl hydroxylase 2 (17%). The best predictors for MAP kinase kinase 2 were squalene epoxidase (59%), kinesin family member C1 (30%), and cellular retinoic acid binding protein II (19%). Matrix metalloproteinase-2 was best predicted by Stat1 (27%), guanidinoacetate methyltransferase (18%), Epstein-Barr virus-induced gene 3 (14%), and thioredoxin peroxidase (13%). A large number of predictor sets were identified for cyclin D gene with class I-L region (12%), histocompatibility class I-K region (11%), and galectin-9 (10%) among the most frequently encountered. Because so many different clones were found to predict cyclin D, no one gene predominated.
Relevant BaP targets (such as the major histocompatibility class I-L region, IGFBP-5, lymphocyte antigen 6 complex, and thioredoxin peroxidase) shared several common predictors. MSSP was identified as a powerful predictor for histocompatibility class I-L region (48%), thioredoxin peroxidase (16%), and lymphocyte 6 antigen complex (22%). Squalene epoxidase (28%), interferon-induced guanylate binding protein GBP-2 (13%), and mitogen-activated protein kinase (10%) were identified as predictors of IGFBP-5. In the case of histocompatibility class I-L region, 6-phosphofructo-2-kinase (52%) and RNA binding motif protein 14 (33%) were identified as frequent predictors. Y box protein (25%), and MSSP (22%), and CD6 antigen (21%) were identified as strong predictors of lymphocyte 6 antigen. For thioredoxin peroxidase, interferon-activated gene 20 (32%), glutamate oxaloacetate transaminase (24%), and B-cell leukemia/lymphoma 3 (24%) were identified as the most frequent predictors.
 |
Discussion
|
---|
The present study focused on the identification of adaptive gene sets during the early phases of BaP-induced atherogenesis. Independent grouping methods identified several gene regulatory networks influenced by hydrocarbon challenge in a redox-dependent manner. At the first level of analysis, ANOVA provided a convenient tool for assessment of significant differences among multiple biological samples. This statistical approach proved superior to simple rules that evaluate fold changes in gene expression because it allowed for inclusion of genes that exhibited small fold changes, but showed significant differences based on a high degree of precision between replicates. Likewise, genes that exhibited a large fold change in one array but high variability across multiple arrays were rendered statistically insignificant and therefore were ignored in subsequent analysis. ANOVA was combined with clustering to uncover group structures within the dataset and identify genes with similar expression patterns (9, 16). This approach significantly reduced the number of genes under study and readily identified patterns of redox sensitivity. Although the complexity of our experimental design precluded analysis of multiple BaP concentrations or durations of exposure, the spectrum of biochemical and molecular changes elicited by BaP in cultured murine VSMCs is known to be both concentration dependent and time dependent (20, 21). As such, the genomic profiles observed are representative of the adaptive stress response, as opposed to transient changes associated with the acute response to hydrocarbon injury.
All clustering methods readily identified genes altered by atherogenic challenge in a redox-specific manner. Genes that clustered together regardless of the method of analysis applied included histocompatibility class I-K region, histocompatibility class I-L region, histocompatibility class I-Q region locus, and HMG-I(Y). Genes known to be modified by oxidative injury, such as PDGF and IGF (24), were identified as redox-sensitive targets, as well as a large number of interferon-related genes involved in the fibro-proliferative inflammatory response. The specificity of the VSMC response was not directly addressed in this study, but recent evidence indicates that treatment of cultured murine metanephric kidney with 3 µM BaP is associated with kidney-specific repression of developmentally regulated genes (glial-derived neurotrophic factor, frizzled receptor, and kidney androgen regulated protein) (Falahatpisheh et al., unpublished data, 2003). Also relevant within this context is the finding that a common set of redox-sensitive target genes is modified within the aortic wall of Sprague-Dawley rats subjected to repeated in vivo cycles of oxidative stress (Partridge CR, Williams ES, Lu KP, Chao S, Johnson CD, Mouneimni R, Meininger GA, Wilson E, and Ramos KS, unpublished observations). A number of genes, however, were specific to BaP, suggesting that gene activation and repression in murine aortic SMCs involve some degree of specificity.
Several matrix genes (procollagen type III-
and tissue inhibitor of metalloproteinase-3) were significantly altered by BaP. This finding is consistent with reports showing that matrix deposition within the vascular wall is influenced by redox status (45) and that the type of matrix surrounding VSMCs influences their phenotypic status (46). For instance, type I and III fibrillar collagen predominate in healthy arteries, whereas proteoglycans intermixed with scattered collagen fibrils predominate in atherosclerotic lesions. Other matrix types, such as fibronectin and heparan sulfate, participate in cell-matrix regulation by macrophage-derived cytokines during cycles of vascular injury (35). Matrix changes are not only important for remodeling of the vascular wall during cycles of injury but also influence recruitment of inflammatory cells. In this context, the regulation of histocompatibility class I-K, -L, and -Q region locus and complement component factors by BaP is of particular interest since molecules associated with leukocyte migration across the endothelium, in conjunction with monocyte chemotactic protein 1, osteopontin, and modified LDL, attract monocytes and T cells into the artery wall (15). Furthermore, interferon-inducible guanylate binding protein GBP-2, IIGP protein, and lymphocyte antigen 6 complex participate in the T-cell activation cascade.
Free radicals and oxidized LDL are cytotoxic to macrophages, VSMCs, and endothelium (43, 44), with both necrosis and apoptosis contributing to formation of the central necrotic core in advanced lesions (1, 18, 19). The balance of cell death and proliferation is important in atherosclerosis, with proliferation, necrosis, and apoptosis all enhanced in atherogenic VSMC phenotypes (25). This balance is altered by interferon in a redox-dependent manner (24), a pattern that further implicates interferon-regulated genes as targets of BaP within the vascular wall. Together, these findings suggest that inflammatory cell activation and recruitment are part of the sequence of events leading to atherogenesis in response to BaP treatment.
The interaction between circulating lymphocytes, endothelium, and VSMCs is a crucial step in atherogenesis (43). T lymphocytes are among the earliest cells infiltrating the arterial intima during the initial stages of atherosclerosis and are commonly found in fatty streaks along with monocytes. Activated CD8+ T lymphocytes express interferon-
(IFN-
), and nearly all cells express receptors for IFN-
and respond to ligand binding by increasing surface expression of major histocompatibility complex class I proteins and promoting the presentation of antigen to T-helper (CD4+) cells. IFN-
also increases presentation of major histocompatibility complex class II proteins, further enhancing the ability of cells to present antigen to T cells and increasing the overall inflammatory response. Class I major histocompatibility complex-deficient mice demonstrate a threefold increase in atherosclerotic lesion area (13). Thus our findings implicate BaP as an inducer of genomic changes in VSMCs that modify T-cell activation profiles. Secreted phosphoprotein, an osteopontin precursor and part of the early T-cell recognition complex, was also highly regulated by BaP. This gene was sensitive to alterations in redox status, suggesting that the balance between pro-inflammatory and anti-inflammatory cytokines may be a decisive factor for the induction of atherogenic phenotypes by BaP. Treatment of VSMCs with BaP also induced several growth factors, such as IGF and PDGFr-
, providing further evidence for the atherogenic potential of this hydrocarbon.
A novel computational target clone-prediction analysis was used to unravel biological networks during atherogenesis. Many of the predictor genes identified, such as c-myc single-strand binding protein (MSSP), fit well within our current understanding of oxidant-induced atherosclerosis (12). BaP modulates c-myc expression in VSMCs (42), and MSSP is a modulator of c-myc and cellular transformation (33). The finding that an apoptosis gene may be predicted by the behavior of a DNase inhibitor would likewise be expected. Other critical relationships found included the interaction between thioredoxin peroxidase and apoptosis. Studies have shown that thioredoxin peroxidase acts as an inhibitor of apoptosis in a mechanism distinct from that of Bcl-2 (47). These well-established relationships between predictors and targets lend support to unanticipated relationships, such as the putative linkage between fyn proto-oncogene, Ndr3, and hemochromatosis gene as predictors of the Ahr. Work in this laboratory has shown that activation of Ahr by BaP is critical for the deregulation of growth-related gene expression in VSMCs (20). One of the most intriguing predictor/target relationships was the emergence of histocompatibility complex associated genes as predictors of cyclin D and lysyl oxidase. Cyclin D is intimately involved in cell cycle control, while lysyl oxidase functions in maturation of collagen and elastin and as a putative tumor suppressor through a Ras-related mechanism (5). Ras is important in VSMC differentiation, cell cycle regulation, and apoptosis (39).
Collectively, the data presented here indicate that induction of atherogenic phenotypes by BaP involves redox-dependent modulation of gene expression. Critical molecular targets during the early adaptive response include genes involved in cell cycle control, matrix deposition, and apoptosis. Among the most intriguing findings was the identification of immune-related genes that modify recruitment and activation of T cells and monocytes (lymphocyte antigen-6 complex, histocompatibility class I component factors, secreted phosphoprotein, and several interferon-inducible proteins) as critical molecular targets during the early stages of atherogenesis. Although treatment of murine VSMCs with BaP is not representative of all forms of oxidative stress encountered by the arterial wall, our findings yield insight into the molecular basis of oxidant-induced atherogenesis and open new venues of exploration to those interested in VSMC pathobiology caused by oxidative stress.
 |
ACKNOWLEDGMENTS
|
---|
This work was supported by National Institutes of Health Grants ES-04849, ES-09106, and ES-07273 (to K. S. Ramos) and CA-90301 (to R. J. Carroll).
 |
FOOTNOTES
|
---|
Address for reprint requests and other correspondence and present address for K. S. Ramos: Dept. of Biochemistry and Molecular Biology, Univ. of Louisville Health Sciences Center, Louisville, KY 40292 (ksramo01{at}gwise.louisville.edu).
10.1152/physiolgenomics.00006.2003.
1 The Data Supplement for this article (APPENDIX 1 and APPENDIX 2) is available online at http://physiolgenomics.physiology.org/cgi/content/full/13/3/263/DC1. 
 |
References
|
---|
- Bennett MR. Apoptosis of vascular smooth muscle cells in vascular remodeling and atherosclerotic plaque rupture. Cardiovasc Res 41: 361368, 1999.[ISI][Medline]
- Chen Y, Kamat V, Dougherty ER, Bittner ML, Meltzer PL, and Trent JM. Ratio statistics of gene expression levels and applications to microarray data analysis. Bioinformatics 8: 12071215, 2002.
- Chen Y, Dougherty ER, and Bittner ML. Ratio based decisions and the quantitative analysis of cDNA microarray images. J Biomed Optics 2: 364374, 1997.
- Chen YH and Ramos KS. A CCAAT/enhancer binding protein (C/EBP) site within antioxidant/electrophile response element (ARE/EpRE) along with CREB binding protein (CBP) participate in the negative regulation of rat GST-Ya gene in vascular smooth muscle cells. J Biol Chem 275: 2736627376, 2000.[Abstract/Free Full Text]
- Contente S, Kenyon K, Rimoldi D, and Friedman RM. Expression of gene rrg is associated with reversion of NIH 3T3 cells by LTR-c-Ha-ras. Science 249: 796798, 1990.[ISI][Medline]
- DeRisi J, Iyer V, and Brown P. Exploring the metabolic and genetic control of gene expression on a genomic scale. Science 278: 680686, 1997.[Abstract/Free Full Text]
- Dougherty ER, Barrera J, Brun M, Kim S, Cesar RM, Chen Y, Bittner M, and Trent JM. Inference from clustering with application to gene-expression microarrays. J Comput Biol 9: 105126, 2002.[ISI][Medline]
- Dougherty ER, Kim S, and Chen Y. Coefficient of determination in nonlinear signal processing. Signal Processing 80: 22192235, 2000.[ISI]
- Eisen M, Spellman P, Brown P, and Botstein D. Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci USA 95: 1486314868, 1998.[Abstract/Free Full Text]
- Fan YY, Ramos KS, and Chapkin RS. Dietary
-linolenic acid suppresses aortic smooth muscle cell proliferation and modifies atherosclerotic lesions in apolipoprotein E knockout mice. J Nutr 131: 16751681, 2001.[Abstract/Free Full Text]
- Fisher RA. The Design of Experiments. Edinburgh: Oliver and Boyd, 1935.
- Fujimoto M, Matsumoto K, Iguchi-Ariga SM, and Ariga H. Disruption of MSSP, c-myc single-strand binding protein, leads to embryonic lethality in some homozygous mice. Genes Cells 6: 10671075, 2001.[Abstract/Free Full Text]
- Fyfe AI, Qiao J, and Lusis AJ. Immune-deficient mice develop typical atherosclerotic fatty streaks when fed an atherogenic diet. J Clin Invest 94: 25162520, 1994.[ISI][Medline]
- Gelboin HV. Benzo[a]pyrene metabolism, activation and carcinogenesis: role and regulation of mixed-function oxidases and related enzymes. Physiol Rev 60: 11071066, 1980.[Free Full Text]
- Hansson GK. Immune mechanism in atherosclerosis. Arterioscler Thromb Vasc Biol 21: 18761890, 2001.[Abstract/Free Full Text]
- Hochberg BY. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J Stat Soc B 57: 289300, 1995.[ISI]
- Howard G, Wagenknecht LE, Burke GL, Diez-Roux A, Evans GW, McGovern P, Nieto FJ, and Tell GS. Cigarette smoking and progression of atherosclerosis: The Atherosclerosis Risk in Communities (ARIC) Study. JAMA 279: 119124, 1998.[Abstract/Free Full Text]
- Isner JM, Kearney M, Bortman S, and Passeri J. Apoptosis in human atherosclerosis and restenosis. Circulation 91: 27032711, 1995.[Abstract/Free Full Text]
- Kalai M, Van LG, Vanden BT, Meeus A, Burm W, Saelens X, and Vandenabeele P. Tipping the balance between necrosis and apoptosis in human and murine cells treated with interferon and dsRNA. Cell Death Differ 9: 981994, 2002.[ISI][Medline]
- Kerzee JK and Ramos KS. Activation of c-Ha-ras by benzo(a)pyrene in vascular smooth muscle cells involves redox stress and aryl hydrocarbon receptor. Mol Pharmacol 58: 152158, 2000.[Abstract/Free Full Text]
- Kerzee JK and Ramos KS. Constitutive and inducible expression of CYP1A1 and CYP1B1 in vascular smooth muscle cells: role of the AHR bHLH/PAS transcription factor. Circ Res 89: 573582, 2001.[Abstract/Free Full Text]
- Kim S, Dougherty ER, Bittner ML, Chen Y, Sivakumar K, Meltzer PS, and Trent JM. General nonlinear framework for the analysis of gene interaction via multivariate expression arrays. J Biomed Optics 5: 411424, 2000.[ISI]
- Kim S, Dougherty ER, Bittner ML, Chen Y, Sivakumar K, Meltzer PS, and Trent JM. Multivariate measurement of gene expression relationships. Genomics 67: 201209, 2000.[ISI][Medline]
- Klahr S and Morrissey JJ. The role of vasoactive compounds, growth factors and cytokines in the progression of renal disease. Kidney Int Suppl 75: S7S14, 2000.[Medline]
- Kockx MM. Apoptosis in the atherosclerotic plaque: quantitative and qualitative aspects. Arterioscler Thromb Vasc Biol 18: 15191522, 1998.[Abstract/Free Full Text]
- Li Z, Khaletskiy A, Wang J, Wong JY, Oberley LW, and Li JJ. Genes regulated in human breast cancer cells overexpressing manganese-containing superoxide dismutase. Free Radic Biol Med 30: 260267, 2001.[ISI][Medline]
- Lu KP, Alejandro NF, Taylor KM, Joyce MM, Spencer TE, and Ramos KS. Differential expression of ribosomal L31, Zis, gas-5 and mitochondrial mRNAs following oxidant induction of proliferative vascular smooth muscle cell phenotypes. Atherosclerosis 160: 273280, 2002.[ISI][Medline]
- Lu KP and Ramos KS. Identification of genes differentially expressed in vascular smooth muscle cells following benzo(a)pyrene challenge: implications for chemical atherogenesis. Biochem Biophys Res Commun 253: 828833, 1998.[ISI][Medline]
- Lu KP, Hallberg LM, Tomlinson J, and Ramos KS. Benzo(a)pyrene activates L1Md retrotransposon and inhibits DNA repair in vascular smooth muscle cells. Mutat Res 454: 3544, 2000.[ISI][Medline]
- Miller KP and Ramos KS. Impact of metabolism on the biological effects of benzo(a)pyrene and related hydrocarbons. Drug Metab Rev 33: 135, 2001.[ISI]
- Moorthy B, Miller KP, and Ramos KS. Metabolic activation of benzo[a]pyrene by CYP1B1 in mouse vascular smooth muscle cells: evidence from 32P-postlabeling analysis for formation of 3-hydroxybenzo[a]pyrene and benzo[a]pyrene-3,6-quinone as major proximate genotoxic intermediates. J Pharmacol Exp Ther 305: 394401, 2003.[Abstract/Free Full Text]
- Nebert DW, Puga A, and Vasiliou V. Role of the Ah receptor and the dioxin-inducible [Ah] gene battery in toxicity, cancer, and signal transduction. Ann NY Acad Sci 685: 624640, 1993.[Abstract]
- Niki T, Izumi S, Saegusa Y, Taira T, Takai T, Iguchi-Ariga SM, and Ariga H. MSSP promotes ras/myc cooperative cell transforming activity by binding to c-Myc. Genes Cells 5: 127141, 2000.[Abstract/Free Full Text]
- Ou X and Ramos KS. Proliferative responses of quail aortic smooth muscle cells to benzo[a]pyrene: implications in PAH-induced atherogenesis. Toxicology 74: 243258, 1992.[ISI][Medline]
- Palombo D, Maione M, Cifiello BI, Udini M, Maggio D, and Lupo M. Matrix metalloproteinases. Their role in degenerative chronic diseases of abdominal aorta. J Cardiovasc Surg (Torino) 40: 257260, 1999.[ISI][Medline]
- Raines EW and Ross R. Smooth muscle cells and the pathogenesis of the lesions of atherosclerosis. Br Heart J 69: S30S37, 1993.[Medline]
- Ramos KS and Cox LR. Aortic endothelial and smooth muscle cell cultures. In Methods in Toxicology: In Vitro Biological Systems. San Diego, CA: Academic, 1993, p. 159168.
- Ramos KS, Zhang Y, Sadhu DN, and Chapkin RC. The induction of proliferative phenotypes in vascular smooth muscle cells by benzo[a]pyrene is characterized by upregulation of phosphatidylinositol metabolism and c-Ha-ras gene expression. Arch Biochem Biophys 332: 213222, 1996.[ISI][Medline]
- Ramos KS. Redox regulation of c-Ha-ras and osteopontin signaling in vascular smooth muscle cells. Implications in chemical atherogenesis. Annu Rev Pharmacol Toxicol 29: 243265, 1999.
- Ross R. Atherosclerosis an inflammatory disease. N Engl J Med 340: 115126, 1999.[Free Full Text]
- Sadhu DN, Merchant M, Safe SH, and Ramos KS. Modulation of protooncogene expression in rat aortic smooth muscle cells by benzo[a]pyrene. Arch Biochem Biophys 300: 124131, 1993.[ISI][Medline]
- Schwartz CJ, Valente AJ, Sprague EA, Kelley JL, and Nerem RM. The pathogenesis of atherosclerosis: an overview. Clin Cardiol 14: I1I16, 1991.[Medline]
- Shi X, Ding M, Dong Z, Chen F, Ye J, Wang S, Leonard SS, Castranova V, and Vallyathan V. Antioxidant properties of aspirin: characterization of the ability of aspirin to inhibit silica-induced lipid peroxidation, DNA damage, NF-
B activation, and TNF-
production. Mol Cell Biochem 99: 93102, 1999.
- Uemura S, Matsushita H, Li W, Glassford AJ, Asagami T, Lee KH, Harrison DG, and Tsao PS. Diabetes mellitus enhances vascular matrix metalloproteinase activity: role of oxidative stress. Circ Res 88: 12911298, 2001.[Abstract/Free Full Text]
- Wilson E, Parrish AR, Bral CM, Williams ES, and Ramos KS. Collagen suppresses the proliferative phenotype of allylamine-injured vascular smooth muscle cells. Atherosclerosis 162: 289197, 2002.[ISI][Medline]
- Zhang P, Liu B, Kang SW, Seo MS, Rhee SG, and Obeid LM. Thioredoxin peroxidase is a novel inhibitor of apoptosis with a mechanism distinct from that of Bcl-2. J Biol Chem 272: 3061530618, 1997.[Abstract/Free Full Text]