1 Pharmacology Department
2 Drug Metabolism Department, Merck Research Laboratories, West Point, Pennsylvania 19486
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() |
---|
DNA microarray; drug metabolism; gene regulation; cytochrome P-450
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() |
---|
Metabolism of xenobiotics and steroid hormones is carried out in three "phases" (12). Drug-metabolizing enzymes modify substrates directly via oxidation, hydroxylation, and dealkylation (phase I enzymes), by conjugating drugs to sulfate, glutathione, or carbohydrates to facilitate clearance (phase II enzymes), and by directional transport of these conjugates from liver into the bile or urine (phase III enzymes). We focused initially on the cytochromes P-450, including CYP3A, CYP2D, CYP2C, and CYP4A family members, that collectively metabolize 90% of commercial drugs (7). Since genetic variations, polymorphisms, are common in several of the important CYP genes in human populations, it is important to be able to monitor regulation of these enzymes in clinical subjects. The presence of inactive alleles can lead to persistence of drugs in susceptible subjects leading to potential adverse reactions. Inactive alleles in CYP2C9, for example, lead to persistence and life-threatening toxicity of warfarin levels in serum. (13) Thus it is important to understand which enzymes metabolize a drug candidate.
A drug that is metabolized by a given cytochrome P-450 frequently induces expression of that P-450 at the transcriptional level (34). Induction of particular cytochromes P-450 in the liver can herald either harmful or benign clinical events. Although existing enzymatic assays and mass spectrometric analyses are necessary to determine which gene subfamily is actually metabolizing a given compound, these assays often fail to distinguish gene subfamily members. DNA microarray assays are thus complementary to enzymatic assays, as they help provide a prediction of which genes are metabolizing the drug and a facile overview of the livers responses to the drug. DNA microarray assays could also be used to compare transcriptional changes caused by drugs to identify therapeutic mechanisms and potential side effects (20). If a compound causes an adverse effect, then transcriptional changes in the liver may provide clues to the mechanism of the toxic insult. Such insults may be oxidative, tumor initiating or promoting, or inflammatory, for example.
DNA microarray technology typically encompasses arraying of DNA on a glass surface, hybridization and detection methods, and software and databases that facilitate data analyses. DNA microarrays representing thousands of genes can be hybridized to fluorescence-labeled samples representing mRNA populations from control and experimental tissues (11). Expression of each surveyed gene is detected and quantitated as fluorescence from the hybridizing nucleic acid. It should be noted that such data represent steady-state mRNA abundance, reflecting both changes in transcription rate and rate of message destruction. Differentially expressed genes can be identified by informatic comparison of control and experimental hybridization data. Such data can be maintained in relational databases to allow queries to be performed on any or all data simultaneously.
Effective methods have been developed for spatially addressed synthesis of short oligonucleotides, or mechanical arraying of cDNA inserts, on derivatized glass surfaces (10, 27). Synthetic oligonucleotides allow rational avoidance of DNA sequences that are shared by several genes or repeated in the genome. The 25-mer oligonucleotide GeneChip microarray format of Fodor et al. (10) was used in the studies described, due to its potential for discriminating between the closely related gene sequences prevalent in drug metabolism gene families.
Interpretation of complex transcriptional responses to xenobiotics and subsequent prediction of physiological effects is currently problematic. Drawing links between gene expression profiles and physiological effects demands formation of a database using compounds with well-characterized effects from a variety of mechanistic classes. To demonstrate the feasibility and validity of this approach, we examined the responses of drug metabolism genes in the rat liver to four well-studied compounds [3-methylcholanthrene (3MC), dexamethasone, phenobarbital, and clofibrate] using DNA microarrays.
![]() |
MATERIALS AND METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() |
---|
Rat treatment protocols.
Sprague-Dawley rats were purchased from Charles River Laboratories, Wilmington, MA. Rats were watered and fed chow ad libitum. Animal treatments were reviewed and approved by the Institutional Animal Care and Use Committee. Rats were dosed orally once per day for 3 days with 0.1% methylcellulose vehicle or 30 mg·kg-1·day-1 of 3MC, dexamethasone, phenobarbital, or clofibrate. On day 4, rats were killed, and livers were harvested and frozen immediately on dry ice. mRNA was isolated from individual livers of at least two rats per treatment.
RNA preparation.
Tissues were homogenized in Trizol (Life Technologies, Gaithersburg, MD) using a Polytron (Omni International, Warrenton, VT), and total RNA was isolated from each sample according to the manufacturers instructions. Total RNA was reprecipitated using RNAmate (Biochain, San Leandro, CA); mRNA was isolated using oligo-dT-decorated latex beads (Qiagen, Hilden, Germany) according to the manufacturers instructions.
Hybridization and staining.
Hybridization samples were prepared according to Affymetrix instructions as described (18). Briefly, a primer encoding the T7 RNA polymerase promoter linked to oligo-dT17 was used to prime double-stranded cDNA synthesis from each mRNA sample using Superscript II RNase H- reverse transcriptase (Life Technologies, Rockville, MD). Each double-stranded cDNA sample was purified by adsorption to silica (Qiaquick kit, Qiagen) according to manufacturers instructions, then in vitro transcribed using T7 RNA polymerase (MEGAscript T7 kit; Ambion, Austin, TX), incorporating biotin-UTP and biotin-CTP (Enzo Biochemicals, New York, NY) into the resulting copy RNA (cRNA). These cRNA transcripts were purified using RNEasy (Qiagen) and quantitated by measuring absorption at 260 nm/280 nm. Five-microgram mRNA samples typically yielded between 30 and 150 µg purified cRNA. cRNA samples were fragmented at 95°C for 35 min in 10 mM MgCl2 to a mean size of 50100 nucleotides, added to hybridization buffer, and hybridized to the Merck Drug Safety Chip for 16 h at 45°C. GeneChips were washed, stained with streptavidin-R-phycoerythrin, and scanned with a dedicated instrument to capture a fluorescence image (Molecular Dynamics).
Hybridization spikes.
Biotinylated RNA samples were prepared from individual CYP genes to spike into hybridization samples at known concentrations, as follows. CYP3A gene expressed sequence tag (EST) clones that contained the entire 1,000 nt 3' region represented on the GeneChips were selected. GenBank accession numbers for the CYP3A clones were as follows: CYP3A4, no. T60335; CYP3A5, no. AA740526; and CYP3A7, no. AA455159. The CYP gene and the cDNA clones in vector pCDNA3.1 were obtained from the following: Frank Gonzalez, National Institutes of Health, supplied CYP2A6, GenBank no. M33318; Joyce Goldstein National Institute of Environmental Health Sciences, supplied CYP2C19, GenBank no. M61854; and Tom Rushmore supplied CYP2D6, which is unpublished. The inserts from CYP3A4 (in pBluescript SK-; Stratagene, La Jolla, CA), CYP3A5, and CYP3A7 (EST clones in pT7T3D-Pac; Ref. 30) or full-length clones in pCDNA3.1 (Invitrogen, Carlsbad, CA) were amplified by PCR using primers that match DNA sequences flanking the cDNA insertion sequences in each vector. PCR primers were the "T7" primer (5' TAATACGACTCACTATAGGG) and a "T3-pCDNA3.1" primer (for pCDNA3.1: 5' AGATGCAATTAACCCTCACTAAAGGGAGAGAAGGCACAGTCGAGGCTGA) or a "T3" primer (for EST clones: 5' ATTAACCCTCACTAAAGGGA) that encodes a T3 RNA polymerase promoter sequence. PCR products were purified by adsorption to silica (Qiaquick kit, Qiagen) according to the manufacturers instructions. Each PCR product was then transcribed in vitro to generate the antisense strand of each gene using T3 RNA polymerase for pCDNA3.1 and pT7T3D-Pac or using T7 RNA polymerase for pBluescript SK-, incorporating biotin-UTP and biotin-CTP into the resulting cRNA. These cRNA transcripts were purified using RNEasy (Qiagen), quantitated by measuring absorption at 260 nm/280 nm, and spiked into hybridizations at known concentrations.
GeneChip data analysis.
Data from each microarray were normalized to data from a single vehicle microarray using global scaling based on the overall hybridization intensities. Normalization, assessments of replicates, and calculations of gene expression levels as average difference values were performed using GeneChip v. 3.1 software. Each treatment was represented by two replicate samples, using three pooled rat livers per sample and two microarrays.
Quantitative RT-PCR.
TaqMan quantitative reverse transcriptase PCR (Q-RT-PCR) was used to quantitate mRNA levels for selected genes as described in detail by Wang and Brown (33). Two primers and one probe were designed for each gene using Perkin-Elmer PrimerExpress v. 1.0 software.
Primer and probe sequences are listed for forward primers "_F", reverse primers "_R", and probes "_M" which were labeled with 6-carboxyfluorescein (FAM) on the 5' nucleotide and 6-carboxy-N,N,N',N'-tetramethylrhodamine (TAMRA) on the 3' nucleotide.
APOA1_F GTTTGGGCTCTACAGCGATCAG
APOA1_R TGGTTCTTGATCTCGGTCAGG
APOA1_M TGCGCGAGAACCTGGCCCA
BE_F CCTCTGAAGGAATGGCAAAGC
BE_R CGCAGAAGGCTGAATCACAG
BE_M CAGGGCCCCACGGCAGCA
CYP1A1_F GGAGGTGCTCTTGCCATCTG
CYP1A1_R CACCCCCACATGTAGTGTCATAA
CYP1A1_M TGAGGCTCAACTGTCTTCCAACATGGG
CYP3A18_F TGGCATGAGGTTTGCTCTCA
CYP3A18_R TCACAAGGCTGGATATTGAAGTTC
CYP3A18_M CAGCATGAAACTTGCTGTCATAGGAGTCCTG
GSTM1_F AGACAGAGGAGGAGCGGATTC
GSTM1_R GCTGCATGCGGTTGTCC
GSTM1_M TGCCGACATTGTGGAGAACCAGGTC
Assays were run on the Perkin-Elmer ABI 7700 instrument under default conditions. Abundance of each gene was determined relative to a standard transcript, 18S rRNA. cDNA was prepared from two pools of three livers per treatment group, and each cDNA was assayed in duplicate PCR reactions.
Quantitative RT-PCR data analysis.
Average Ct values from duplicate PCR reactions were normalized to average Ct values for 18S rRNA from the same cDNA preparations. The ratio of expression of each gene in drug-treated vs. vehicle-treated samples was calculated as 2-(meanCt) of that treatment as recommended by Perkin-Elmer where Ct is the threshold cycle, and
Ct is the difference Ct(test gene) - Ct(18S rRNA) for treated sample minus vehicle sample. Using the ANOVA method, 95% confidence intervals were determined for each ratio as
![]() |
![]() |
RESULTS AND DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() |
---|
Despite this redundancy, some experimental variation is observed in gene expression levels even in replicate hybridizations of the same sample. Such variation can lead to identification of "false positives." We use the following empirical cutoff values to screen out false positives: the absolute expression levels (average difference values) between two hybridizations differ by 20 units and by a ratio of
2-fold. We derived these criteria from previous experiments in which pairs of hybridization samples were derived from a common mRNA sample and hybridized to replicate microarrays. In these experiments, the false-positive rate ranged from 0.41.8% of genes detected using these criteria (unpublished observations). If false-positive events in different hybridization samples arise independently, then two replicates of each sample would yield a false-positive rate of 0.16 to 4 per 10,000. Thus replication of samples is an efficient method by which to identify false positives that arise independently. Experiments in which multiple samples were analyzed from each single liver indicate that interindividual biological variation exceeds experimental variation in this biological system. Hence, three livers were pooled for analysis on each microarray. Such data also suggest that a subset of genes detected are hypervariable between individuals, although very large data sets will be required to define levels of interindividual variation and assign levels of variability to particular genes.
Gene specificity.
We tested the ability of the Affymetrix technology to discriminate between closely related genes using the large CYP gene family. DNA homology relationships between the 3' 1,000 bp of some human cytochromes P-450 examined in this study are depicted in Fig. 1. Probe sequences representing each gene were selected for predicted hybridization efficiency and filtered to exclude probe sequences that are complementary to >23 of 25 bases in other gene sequences represented on the same microarray. Probe specificity was evaluated using a microarray designated "Merck-1" that represents each of 32 human CYP genes by 60 probe pairs from the 3'-terminal 1,000 bases of each gene sequence. Biotin-labeled RNAs were prepared by in vitro transcription of cDNA clones for human genes: CYP2A6, CYP2C19, CYP2D6, CYP3A4, CYP3A5, and CYP3A7. These biotin-RNAs were combined in sets of four to include one CYP3A gene (CYP3A4, CYP3A5, or CYP3A7) per set and hybridized to microarray "Merck-1." Each transcript was used at 30 pM in a 200-µl hybridization sample. Under these conditions, 30 pM corresponds to roughly 1 part in 5,000 in an mRNA population, or about 60 transcripts per mammalian cell.
|
|
The cross-hybridization among CYP3A genes, but not other gene subfamilies with similar homologies (compare Fig. 1 and Fig. 2), was unexpected. The relatively poor gene specificity of probes representing CYP3A5 may have been a consequence of the short 3'-untranslated region (3'-UT) used to design probes for the CYP3A5 gene compared with the long 3'-UT of the CYP3A4 and CYP3A7 genes. This short 3'-UT necessitated that the 1,000 bp region represented on the microarray occur primarily in the coding region of the gene. This region of the gene appears to permit heterologous hybridization despite showing lower sequence identities to CYP3A4 and CYP3A7 (88% and 89%, respectively) than the 92% identity shared by the 3' ends of CYP3A4 and CYP3A7.
Drug metabolism gene regulation.
We further validated the application of DNA microarrays to drug safety by treating rats with four compounds that provoke well-characterized transcriptional responses among the CYP superfamily. 3MC, phenobarbital, dexamethasone, and clofibrate are classic inducers of the cytochrome P-450 subfamily members CYP1A, CYP2B, CYP3A, and CYP4A, respectively (7). Several members of the CYP gene families can be transcriptionally activated by xenobiotics through one of four receptor-dependent mechanisms. 3MC activates transcription of the CYP1 family through the aromatic hydrocarbon (AH) receptor (35), phenobarbital activates transcription of the CYP2B and CYP3A family through the CAR (15), and dexamethasone activates transcription of the CYP3A family through the pregnane nuclear receptor (PXR; Ref. 4), and clofibrate activates transcription of the CYP4A subfamily through the peroxisome proliferator-activated receptor- (PPAR
; Ref. 29).
The results from the induction experiments in which male rats were treated with drug or vehicle are shown in Fig. 3. A regimen of intraperitoneal administration once daily for 3 days and harvest of livers at day 4 was selected, based on prior experience, to represent the steady-state response to compounds. Hybridization samples from two animals per treatment were prepared and hybridized to the Merck Drug Safety Chip. Figure 3 includes all genes that were regulated by any of the compounds in these experiments. The results are presented as the mean expression profiles in livers from rats treated with 3MC, phenobarbital, dexamethasone, or clofibrate. Colors were assigned to data points to reflect twofold induction (red) or suppression (green) with a difference of 20 units in drug-treated samples relative to controls. Additional data points that fail one of these criteria, but that are consistent in replicate data and/or substantiated by the literature are colored light red or light green to avoid possible omissions. Several genes in Fig. 3 are homologous to other rat genes at >90% DNA:DNA identity, including CYP2A1:CYP2A2, CYP2B1:CYP2B2, GSTA1:GSTA2, and HSST1:HSST2. It is possible that data for these genes reflects heterologous hybridization, as suggested by the data in Fig. 2.
|
Induction of several phase II enzymes, UGT1A6, GSTA1, GSTA2, and GSTM1, was also observed in the liver recovered from rats treated with 3MC (Fig. 3). 3MC is known to induce expression of the UDP-glucuronosyl transferase gene UGT1A6 (5) and glutathione-S-transferase gene GSTA1 (25), by an AH receptor-dependent mechanism. The GSTA2 and GSTM1 genes are both regulated (induced) by the CYP1A1 epoxide and hydroxylated metabolite(s) of 3MC. Both genes contain an antioxidant response element (ARE) in their promoter region. This cis-acting element has been shown to be responsive to the diol metabolites of 3MC that can redox cycle and produce a pro-oxidative environment (25). The glutathione-S-transferase genes GSTA2 and GSTM1 (19) were previously observed to be 3MC inducible in cultured rat hepatocytes.
The barbiturate phenobarbital is also a transcriptional inducer of the rat genes CYP2B1, CYP2B2, and CYP3A1 (7). In mouse, phenobarbital induction of Cyp2b10 requires the phenobarbital-response element (PBRE), an enhancer upstream of the Cyp2b10 gene. CAR is a nuclear receptor that interacts with the RXR nuclear receptor to form CAR-RXR heterodimers which bind to the PBRE in response to phenobarbital treatment, suggesting CAR-RXR may play a role in phenobarbital induction of mouse Cyp2b10 (15). The microarray data in Fig. 3 show the induction CYP2B1, CYP2B2, and CYP3A1. We observe a moderate but consistent suppression of CYP2B3 by all four compounds. Little is known about CYP2B3 function, but Yamada et al. (35) have reported expression and drug-mediated induction of CYP2B3 in rat liver. CYP2C7 is also reportedly induced by phenobarbital, but primarily in female rather than in male rats (7). No CAR-RXR response element has been identified in the CYP3A1 promoter region.
Induction of several of the phase II enzymes was observed after treatment with phenobarbital. A significant increase in the specific mRNA for microsomal epoxide hydrolase (EHm), UGT2B1, GSTA1, GSTA2, GSTA3, and GSTM1, was also observed in the liver recovered from rats treated with phenobarbital (Fig. 3). Phenobarbital is known to induce expression of the CYP2B genes by the fore-mentioned CAR-dependent mechanism. No PBRE sequence has been identified in the promoters for any of the phase II enzymes to date. In addition to the phase I and II enzyme induction, a significant increase in the cation transporter, cMOAT, was also observed. Little is known about its regulation in rat liver.
Dexamethasone is a synthetic glucocorticoid mimetic that is known to induce expression of several CYP3A subfamily genes. Recent reports have described the identification and cloning of a nuclear receptor (PXR, the pregnane nuclear receptor) that can bind a variety of chemical structures leading to the induction of CYP3A1 in the liver. This nuclear receptor has been identified in mouse (mPXR), in rat (rPXR), in rabbit (rbPXR) and in humans (PAR, or SXR) (17) (4). In a manner similar to that of the CAR, the PXR can also form heterodimeric complexes with the RXR and bind to a cis-acting sequence in the promoter region of the target gene. PXR-responsive cis-acting regulatory elements have been identified in the promoters of CYP3A genes in each of these species. In agreement with these findings, we observed a marked induction of CYP3A1 and CYP3A18 genes in the liver from rats treated with dexamethasone (Fig. 3). A moderate induction of CYP2B1 and CYP2B2 is also evident, as reported by Ronis et al. (24).
Induction of several of the phase II enzymes was also observed after treatment with dexamethasone. Significant increases in the specific mRNAs for UGT1A6, UGT2B1, GSTA1, GSTA2, GSTM1 and for sulfotransferases PST, HSST1, and HSST2 were detected in the mRNA recovered from the livers of rats treated with dexamethasone (Fig. 3). Dexamethasone is known to induce expression of the CYP3A genes by the aforementioned PXR-dependent mechanism. No cis-acting sequence has been identified in the promoters of any of the phase II enzymes to date. In addition to the phase I and phase II enzyme induction, a significant decrease in the levels of mRNA for sulfotransferase EST and phase III transporter NTCP were also observed. Little is known about their regulation in rat liver.
Clofibrate is a lipid-lowering agent that triggers peroxisome proliferation in rodents. The increase in enzyme expression that accompanies the proliferation of peroxisomes is mediated through a fourth nuclear receptor, PPAR. The peroxisome proliferator compounds are considered agonists of PPAR
. As described previously for the CAR and PXR receptors, the PPAR
also forms heterodimeric complexes with RXR. The complex can bind agonist and interact with the cis-acting sequence [peroxisome proliferator response element (PPRE)] in the promoter of responsive genes. PPAR
is also activated by cellular long-chain fatty acid derivatives, stimulating ß-oxidation of long-chain fatty acids in peroxisomes. Clofibrate and other PPAR
agonists are known to induce members of the CYP4A subfamily. Figure 3 shows transcriptional induction by clofibrate of CYP4A1, CYP4A2, and CYP4A3. Additionally, a small but significant increase in the expression of CYP2B1, CYP2B2, and CYP3A1 was observed.
Induction of several of the phase II enzymes was also observed after treatment with clofibrate. Significant increases in the specific mRNA for UGT1A6 and sulfotransferase PST were also observed in the liver of rats treated with clofibrate (Fig. 3). Clofibrate is known to induce expression of the CYP4A genes by the aforementioned PPAR-dependent mechanism. No cis-acting PPRE sequence has been identified in the promoters of any of the phase II enzymes to date. We also observed a significant decrease in the mRNA for several of the phase II and III enzymes. Significant decreases were seen for GSTA1, GSTM1, GSTM2, and the transporter cMOAT. Little is known about their regulation in rat liver. Each of the four compounds examined directly regulates several drug metabolism genes, through binding to a nuclear receptor or the AH receptor, thus activating the promoters of the genes. Figure 3 shows that each compound also regulates a series of other drug metabolism genes by unknown mechanisms. Many of these are likely to be indirect mechanisms, including competition of the agonized-nuclear receptor for other transcription factors, such as RXR, CBP/p300, and histone acetylases. Such indirect mechanisms may also include primary regulation of a gene product that alters metabolism of the natural ligands for nuclear receptors, altering expression of secondary genes.
Metabolic energy and stress-related gene regulation.
In addition to drug-metabolizing genes, Fig. 3 indicates induction or suppression of genes that indicate toxicological events and genes that regulate sugar and lipid metabolism. 3MC and clofibrate treatments, for example, downregulate lecithin:cholesterol acyltransferase (LCAT), apolipoprotein CI (APOC1), apolipoprotein AIV (APOA4), and phosphoenolpyruvate carboxykinase C (PEPCK). These regulatory events indicate decreased lipid turnover, as well as decreased gluconeogenesis mediated by PEPCK, in the liver. Consistent with this data for 3MC, AH receptor agonists have been found to cause lipid accumulation within hepatocytes in vivo (3) and to regulate lipid metabolism via the AH receptor in cultured fibroblasts (1), although the mechanism remains unclear. Induction of apolipoprotein genes APOA1 and APOA4 by phenobarbital and dexamethasone was also observed (Fig. 3), consistent with increased high-density lipoprotein levels and increased apolipoprotein AI expression in response to these agents in previous reports (9, 23).
Clofibrate is known to globally increase lipoprotein uptake and increase fatty acid ß-oxidation in both peroxisomes and mitochondria in liver (31). The increased lipoprotein uptake is reflected in Fig. 3 by induction of hormone-sensitive lipase (HSLIP). Similarly, decreased lipid levels in serum are suggested by downregulation of apolipoprotein genes APOA1, APOA4, and APOC1. Clofibrate suppression of APOA1 expression is opposite to that in humans, but is consistent with previous studies in male rodents (32). Increased lipid ß-oxidation in response to clofibrate is indicated by induction of peroxisomal genes acyl-CoA oxidase (FACO), bifunctional enzyme (BE), and 3-ketoacyl-CoA thiolase (KCAT); and mitochondrial genes carnitine palmitoyl transferase II (CPT2) and medium chain acyl-CoA dehydrogenase (MCACD).
Induction of mitochondrial delta-aminolevulinate synthase (DALS) indicates increased heme synthesis. Such heme likely supports increased CYP4A enzymes, mitochondrial electron transport proteins, or both. Fatty acid ß-oxidation by peroxisomes and mitochondria is thought to cause oxidative stress by producing reactive oxygen intermediates. Evidence of oxidative stress includes induction of the extracellular Cu-Zn superoxide dismutase (SODxc), and heat shock protein 70 (HSP70), as observed in hepatic oxidative stress caused by CCl4 (28). Induction of the SODxc gene by peroxisome proliferators has not been previously reported, although Yoo et al. (36) report induction of the rat intracellular Cu-Zn superoxide dismutase (SOD1) gene by arachidonic acid.
Figure 3 indicates clofibrate induction of the prostacyclin or prostaglandin I2 receptor (PRCR). Prostaglandin I2 mediates blood vessel dilation and inhibits platelet activation (21). Induction of the PRCR may thus be a marker for the antithrombotic activity of clofibrate (21). Induction of the PRCR appears to be part of a larger shift in the balance of prostaglandin species in clofibrate-treated liver. CYP4A1, CYP4A2, and CYP4A3 genes are induced by clofibrate (Fig. 3) and show varying degrees of -hydroxylase activity on prostaglandins (2) as well as fatty acids. The latter fatty acid
-hydroxylase activity commits fatty acids to peroxisomal degradation. The CYP4F enzymes also can form or degrade various prostaglandins (6, 16), although their activities are not well characterized. Figure 3 shows a trend toward downregulation of the CYP4F1, CYP4F4, and CYP4F6 genes by clofibrate. Thus, although the physiological implications are not clear, it appears likely that regulation of the CYP4A, CYP4F, and PRCR genes by clofibrate may decrease thrombotic activity by altering the prostaglandin balance in rats.
Quantitation of mRNA levels by Q-RT-PCR.
An independent method, TaqMan Q-RT-PCR (33), was used to check changes in mRNA levels for selected genes. Since mRNA levels were determined in each sample relative to an internal control, 18S rRNA, the Q-RT-PCR data is expressed as a ratio of mRNA in treated relative to control samples. This method employs a pair of gene sequence-specific PCR primers to amplify cDNA derived from (rat liver) total RNA. During each PCR cycle, the PCR product is quantitated using a third oligonucleotide "probe." This probe anneals to the PCR product between the two primers and is degraded by the template-dependent exonuclease activity of Taq polymerase. The amount of PCR product is quantitated by the exonucleolytic release of a fluorescent dye linked to the probe. Based on microarray data from Fig. 3, the following genes were selected to check the observed magnitudes of large changes and the reliability of moderate changes: CYP1A1, CYP3A18, CYP4A1, GSTM1, BE, and APOA1. A comparison of quantitation by microarray and by Q-RT-PCR is presented in Fig. 4.
|
For GSTM1, both techniques indicate moderate suppression by clofibrate and moderate induction by the other three compounds. Both techniques also indicate sharp induction of BE by clofibrate, at 41-fold by microarray and 78-fold by Q-RT-PCR. Although control levels of BE were not reliably measured by microarray, Q-RT-PCR indicates a moderate suppression of BE mRNA by dexamethasone. For APOA1, both techniques indicate induction by dexamethasone and phenobarbital, but Q-RT-PCR data were too imprecise to confirm or deny the 0.58-fold suppression of APOA1 by clofibrate indicated by microarray.
Validation of the microarray results for these six genes by Q-RT-PCR showed qualitative agreement overall, yet illustrated fundamental differences between the two techniques. The microarray data failed to register expression of CYP1A1 and BE without strong induction by 3MC, whereas Q-RT-PCR reproducibly detects all five genes in all samples (Fig. 4). The sensitivity of the microarray technology allows detection of gene transcripts down to 1 part in 300,000 (18). This "detection threshold" can account for the lower ratios of induction indicated by microarray data for 3MC induction of CYP1A1 and for clofibrate induction of BE. This threshold may also account for moderate differences in apparent induction ratios of CYP3A18 by dexamethasone, CYP4A1 by clofibrate, and GSTM1 by phenobarbital. Alternatively, some of these differences may have resulted from imprecision in the Q-RT-PCR. Microarray technology has the principle advantage over Q-RT-PCR of examining expression of hundreds or thousands of genes in each experiment (18). Thus the microarray can be used to search many genes for those that respond to a stimulus, and Q-RT-PCR can be used in a complementary way, to confirm and extend observations on a few selected genes.
Conclusions.
We have evaluated the microarray technology for use in determining how the rat liver responds to drug treatments at the mRNA level. The technology allowed discrimination of individual genes within cytochrome P-450 gene subfamilies up to 90% DNA identities. This discrimination was accomplished by rational selection of oligonucleotide probes that were gene specific. Three genes from the CYP3A subfamily showing
90% DNA identities were not fully distinguished. A subset of the 25-mer probe pairs from CYP3A5 were identified that show CYP3A5 gene-specific hybridization relative to CYP3A4 and CYP3A7. This combination of rational sequence selection and empirical hybridization analysis enabled us to unambiguously identify a subset of oligonucleotide probes that are gene specific for the most problematic gene, CYP3A5.
Responses of rat genes relevant to drug metabolism and drug safety were measured using microarray technology and four much studied xenobiotic compounds. These gene expression events concur with numerous previous observations recorded in the literature and concur qualitatively with Q-RT-PCR assays from the same study, thus validating application of microarray technology to drug metabolism and drug safety. Several novel gene regulation responses were also recorded, including induction of markers that elucidate toxicological responses such as HSP70 and SODxc.
Advances in DNA microarray technologies and mammalian genome sequencing will soon allow quantitative assessment of expression profiles of all genes in the selected tissues. The ability to predict phenotypic outcomes from gene expression profiles is currently in its infancy, however, and will require additional bioinformatic tools. Such tools will facilitate information gathering from literature and gene databases as well as integration of expression data with animal physiology studies.
![]() |
ACKNOWLEDGMENTS |
---|
![]() |
FOOTNOTES |
---|
Address for reprint requests and other correspondence: D. Gerhold, Merck, WP26A-3000, Broad St. and Sumneytown Pk., West Point, PA 19486 (E-mail: david_gerhold{at}merck.com).
1 Supplementary Material to this article is available online at http:// physiolgenomics.physiology.org/cgi/content/full/5/4/161/DC1.
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() |
---|