Changes in Gene Expression Induced by Carbamazepine and Phenytoin: Testing the Danger Hypothesis

Abstract

The aromatic anticonvulsants carbamazepine (CBZ) and phenytoin (PHN) are associated with a relatively high incidence of idiosyncratic drug reactions (IDRs). If biomarkers could be found that would predict the risk that a drug candidate would cause IDRs it would significantly decrease the risks associated with drug development. The IDRs associated with CBZ and PHN appear to be immune-mediated. The Danger Hypothesis posits that for something to induce an immune response, it must cause some type of cell damage that ultimately causes up-regulation of co-stimulatory molecules on antigen-presenting cells; without this, the response will be immune tolerance. If the Danger Hypothesis is correct, the ability of a drug or its reactive metabolite to induce cell damage or stress may be related to its risk of causing IDRs. In a parallel study reported elsewhere, we found that major metabolites of these two drugs: 3-OH-CBZ and 4-OH-PHN can be oxidized by peroxidases to phenoxyl free radicals, which could cause oxidative stress by redox cycling. In this study using mRNA microarrays, we found that CBZ and PHN treatment induced changes in mRNA expression in mice. Many of the changes were in genes related to Keap1-Nrf2-ARE signaling pathways and enzymes involved in responding to oxidant stressors and reactive metabolites such as glutathione transferase and heat shock proteins. The similar patterns of genes induced by these two drugs are consistent with the clinical observation that those two drugs exhibit cross-sensitivity. These findings are consistent with the induction of cell stress by CBZ and PHN, most likely due to reactive metabolites. Such changes may represent a danger signal and represent a biomarker of the potential that a drug will cause IDRs; however, different drugs likely cause cell stress by different mechanisms and, therefore, the biomarkers for other drugs would likely be different.

Keywords :

• carbamazepine
• phenytoin
• idiosyncratic drug reactions
• biomarkers
• gene expression
• cell stress
• Keap1-Nrf2-ARE signaling pathways

Abbreviations
AP-1,	=	activator protein 1
PPARα,	=	peroxisome proliferator-activated receptor α
ARE,	=	antioxidant response element
CBZ,	=	carbamazepine
Cp,	=	crossing point
GST,	=	glutathione S-transferase
Gapdh,	=	glyceraldehydes-3-phosphate dehydrogenase
HSP,	=	heat shock protein
IDR,	=	idiosyncratic drug reaction
Keap1,	=	Kelch-like ECH-associated protein 1
MA,	=	Macrophage activators
NF-κ B,	=	nuclear factor κ B
Nrf2,	=	nuclear factor-erythroid 2-related factor 2
3-OH-CBZ,	=	3-hydroxycarbamazepine
4-OH-PHN,	=	4-hydroxyphenytoin
OS/RM,	=	oxidative stressors/reactive metabolites
PCR,	=	polymerase chain reaction
PHN,	=	phenytoin, 5,5-diphenylhydantoin
PP,	=	peroxisome proliferators
ROS,	=	reactive oxygen species
Stat-3,	=	signal transducers and activators of transcription-3
UDPG,	=	uridine diphosphoglucose

INTRODUCTION

Idiosyncratic drug reactions (IDRs) add a significant degree of uncertainty to drug development. IDRs are just as idiosyncratic in animals as they are in humans and so they are not detected in current animal testing. If biomarkers could be found that accurately predicted if a drug candidate has a high IDR potential it would have a significant impact on drug development. There is circumstantial evidence to suggest that many IDRs are caused by reactive metabolites rather than the parent drug and most pharmaceutical companies now actively screen for reactive metabolites; however, there are clearly drugs that form reactive metabolites but are not associated with a significant incidence of IDRs, and it is not yet clear that screening for reactive metabolites has made drugs safer. On the other hand, most people would agree that if a drug candidate forms a large amount of reactive metabolite it is a significant liability.

If reactive metabolites are responsible for IDRs it is not clear how they cause this type of adverse reaction. Many IDRs are immune-mediated and it has been postulated that in order for something to induce an immune response it must cause cell damage, otherwise it will be ignored by the immune system; this is the Danger Hypothesis. Applying this hypothesis to IDRs, in order for a reactive metabolite to induce an immune-mediated IDR it may be necessary for the reactive metabolite to cause cell damage (Seguin and Uetrecht, 2003). This is an attractive hypothesis because it might explain why not all drugs that form reactive metabolites are associated with a significant incidence of IDRS. Furthermore, biomarkers of cell stress might predict the risk that a drug candidate will cause a relatively high incidence of IDRs.

Carbamazepine (CBZ) and phenytoin (PHN) are aromatic anticonvulsants, which are associated with a relatively high incidence of IDRs. The pattern of IDRs is very similar for the two drugs and it is referred to as the aromatic anticonvulsant hypersensitivity syndrome (Shear and Spielberg, 1988; Leeder, 1998). Furthermore, if a patient has an IDR to one of these drugs there is a high probability that they will also have an IDR to the other (incidence of 40–58%) (Hyson and Sadler, 1997). This type of cross-sensitivity is not common between drugs and it suggests that the mechanism of the IDR caused by these two drugs is similar ((Shear and Spielberg, 1988). In parallel studies we demonstrated that the major phenol metabolites of these two drugs are oxidized by peroxidases to free radicals (Lu and Uetrecht, 2008 under revision). In this study we looked for evidence that these two drugs induce mRNA expression changes that may represent biomarkers of IDR risk. If the phenoxyl free radicals are involved, this may take the form of biomarkers of oxidative stress. Therefore, we performed microarray analysis of changes in gene expression in CBZ- and PHN-treated mice to see if there were other changes, especially those associated with cell stress, which might represent a biomarker of IDR potential.

MATERIALS AND METHODS

Materials

CBZ (98% pure) and PHN (99% pure) were purchased from Sigma-Aldrich (Oakville, ON, Canada).

Animals

Animal studies were conducted in accordance with the guidelines of the Canadian Council on Animal Care. All mice used in this study were 6–8-wk-old female C57BL/6 mice, and were supplied by Charles River Canada (Montreal, PQ, Canada). Arriving animals were acclimated for a week before experiments.

mRNA Microarray Study

Female C57BL/6 mice were divided into 8 hr, 24 hr, and 48 hr treatment groups, each with 3–4 animals. For the 8 hr treatment group, mice were treated with CBZ or PHN by gavage and sacrificed 8 hr later. For the 24 hr treatment group, mice were treated with two doses of CBZ or PHN by gavage at 0 hr and 16 hr, and sacrificed at 24 hr. For the 48 hr treatment group, mice were treated with three doses of PHN by gavage at 0 hr, 16 hr, and 40 hr, and sacrificed at 48 hr. The dose was 200 mg/kg/dose for CBZ and 50 mg/kg/dose for PHN. The doses were chosen so that the C_maxs of the parent drugs were similar to those reported in humans (data not shown). The control group for each time point consisted of three animals that were treated with vehicle (water) and sacrificed along with the treated animals for that time point. After sacrifice, livers were immediately cut into slices and submerged in RNAlater solution (Qiagen, Mississauga, ON, Canada).

Samples (∼ 150 mg) were taken from similar parts of the liver to decrease sampling differences, stored at 4°C for 24 hr, and then homogenized for 60 sec using a rotor-stator homogenizer in the appropriate volume of RLT buffer (Qiagen, Mississauga, ON, Canada). The mRNA from each sample was isolated using a commercial RNA extracting column (RNeasy Midi, Qiagen, Mississauga, ON, Canada) according to the manufacture's protocol. The concentration of mRNA was normalized and the quality was determined by gel electrophoresis (1% agarose gel) and the ratio of UV absorption at 260/280 nm was 2.1/1. The mRNA samples were shipped on dry ice to Dr. Chris Bradfield's Lab, University of Wisconsin, where the mRNA changes were determined by microarray analysis.

The mRNA of each treated animal was analyzed with a separate microarray while the mRNA from the three controls for each time point were pooled and analyzed with 1 microarray. The microarrays used for the analysis were produced by the Bradfield lab; for more details on the construction of these microarrays see Thomas et al., 2001. A threshold 2-fold change in gene expression was used as a cutoff. For further details of the procedures and statistical analysis, see Thomas et al. (2001).

mRNA Real-time PCR Study

The mRNA from each mouse treated with CBZ or vehicle, was transcribed to cDNA using Omniscrip RT KIT (Qiagen, Mississauga, ON, Canada) and oligo(dT₁₅) primers (Roche, Laval, Quebec, Canada) and RNAse inhibitor (Roche) according to manufactures' protocols. For relative quantification, mRNA expressions of Cyp2c29, Gstmu1, and Hspa8 were correlated with the mRNA expression of the house-keeping gene Gapdh using a RT-PCR kit (Roche, Laval, Quebec, Canada). cDNA was amplified using Faststart Taq DNA Polyerase and SYBR green (Roche, Laval, Quebec, Canada) in a Light Cycler instrument (Roche, Laval, Quebec, Canada). The quantitative PCR conditions are summarized in the following table (Table 1). Primers used in amplification of Cyp2c29, Gstmu1, Hspa8, and Gapdh are also listed. The final concentration of Mg^{2 +}was 3 mM. The ratio of the Cp (real-time PCR crossing point) values of Cyp2c29, Gstmu1, and Hspa8 to Gapdh of each sample was determined by quantitative PCR analysis.

TABLE 1 Real-time PCR conditions and primers for Cyp2c29, Gstmu1, and Hspa8

Program (Cyp2c29, Gstmu1, Hspa8)	# Cycles	Target temperature (°C)	Incubation time	Temperature transition (°C/sec)	Acquisition mode
Denaturation	1	95	10 min	20	None
PCR(Segment1)	45	95	10 sec	20	None
(Segment2)		58	5 sec	20	None
(Segment3)		72	19 sec	20	Single
Melt(Segment1)	1	95	0	20	None
(Segment2)		65	30 sec	20	None
(Segment3)		98	0	0.20	Cont
Cool	1	40	30 sec	20	None

Cyp2c29 primers: Sense primer: 5′ TTCGCTATGGATCTGGTCGTG; Anti-sense primer: 5′ CACAGGGTGAGCCTTTGGTT; Gstmu1 primers; Sense primer: 5′ CGTATGTTTGAGCCCAAGTGC; Anti-sense primer: 5′ GGTCAGGGAGCCAATGAAGAA. Hspa8 primers: Sense primer: 5′ GCGAAAGCACAAGAAAGACAT; Anti-sense primer: 5′ AAGAGCAGCAAATCCTGAACG. Gapdh primers: Sense primer: 5′ CCATCACCATCTTCCAGGAG; Anti-sense primer: 5′ CCTGCTTCACCACCTTCTTG.

RESULTS

Microarray Study

All microarray data from CBZ- and PHN-treated mice livers are posted on the EDGE (Environment, Drugs and Gene Expression) web site (http://edge.oncology.wisc.edu/login.php). CBZ- and PHN-treated mice shared a similar mRNA regulation profile as determined by microarray analysis. Most noticeable mRNA changes were the induction of certain P450s and other Phase I and II metabolic enzymes. Protective enzymes, such as glutathione S-transferase (GST) and stress proteins, such as heat shock proteins (HSP), were also induced.

We found up-regulation of genes related to oxidative stress and confirmed some of these results by quantitative PCR. For CBZ-treated mice, most of the 24 hr samples after two doses had the greatest mRNA level changes. Even for genes like Cyp2b10, Hspa5, and UDPG transferase 2b34, which had more mRNA induction at the 8 hr timepoint after one dose, the 24 hr timepoint mRNA levels were still relatively high. In PHN-treated animals, significant changes were observed in samples from both the 24 hr (after two doses) and 48 hr (after three doses) timepoints. Other similar studies also chose the 24 hr timepoint for their mRNA studies (McMillian et al., 2005). About 20 out of 30 of the most up-regulated genes from samples of both CBZ- and PHN-treated mice were included in the list in Table 2 and showed a high degree of overlap between the two treatments. Some of those overlapping genes are: Cyp2b10, Cyp2c29, Gstmu1, Cyp1a2, Ugt2b34 and Cyp2a4.

TABLE 2 Selected mRNA changes in CBZ- and PHN-treated mice

		mRNA level change (Fold)
		CBZ		PHN
Accession no. (Gene name)	Description of genes	8 hr	24 hr	8 hr	24 hr	48 hr
NM_009997.1 (Cyp2a4)	Cytochrome P450, family 2, subfamily a, polypeptide 4	(1.7)	(5.6)	(1.4)	(1.5)	(1.8)
		1.37	8.05	1.00	1.57	2.56
		2.07	3.52	1.89	1.49	1.18
		1.67	5.22	1.23		1.66
						1.66
NM_009993.2 (Cyp1a2)	Cytochrome P450, family 1, subfamily a, polypeptide 2	(1.7)	(2.2)	(1.5)	(2.1)	(1.6)
		1.13	2.47	1.17	2.19	1.40
		2.16	1.41	1.54	1.97	1.23
		1.78	2.73	1.70		1.73
						1.96
NM_007815.2 (Cyp2c29)	Cytochrome P450, family 2, subfamily c, polypeptide 29	(5.0)	(7.1)	(3.2)	(5.0)	(4.0)
		3.17	8.68	3.02	5.01	4.11
		6.76	5.43	3.57	4.94	3.59
		5.06	7.22	2.96		3.53
						4.86
NM_009999.2 (Cyp2b10)	Cytochrome P450, family 2, subfamily b, polypeptide 10	(8.0)	(6.8)	(5.7)	(7.1)	(6.3)
		3.10	5.61	5.94	7.09	5.03
		9.95	5.10	5.46	7.02	4.52
		10.9	9.67	5.62		9.79
						5.85
NM_010358.2 (Gstm1)	Glutathione S-transferase, mu 1	(2.5)	(5.0)	(1.2)	(2.5)	(3.4)
		2.27	5.55	−1.20	1.71	4.62
		2.01	3.95	1.47	3.33	2.12
		3.14	5.42	1.42		3.48
						3.20
NM_008181.2 (Gsta1)	Glutathione S-transferase, alpha 1 (Ya)	(−1.1)	(1.8)	(1.0)	(1.4)	(1.2)
		1.01	2.80	−1.09	1.36	1.17
		−1.27	1.58	1.14	1.39	1.23
		1.03	−1.01	1.06		1.38
NM_010356.2 (Gsta3)	Glutathione S-transferase, alpha 3	(1.0)	(1.6)	(1.0)	(91.3)	(1.3)
		1.02	1.33	−1.02	1.47	1.36
		−1.10	2.02	−1.01	1.05	1.50
		1.04	1.41	1.03		1.24
						1.13
NM_008185.2 (Gstt1)	Glutathione S-transferase, theta 1	(1.1)	(1.4)	(1.0)	(1.4)	(1.5)
		1.15	1.06	1.01	1.79	1.29
		1.06	1.98	1.03	1.05	1.83
		1.15	1.29	1.12		1.48
						1.31
(Dgcr2)	Glutathione S-transferase, pi 2	(1.3)	(3.5)	(−1.1)	(1.8)	(1.6)
		1.37	5.07	−1.13	1.43	1.56
		1.50	4.19	−1.11	2.15	1.16
		−1.08	1.11	−1.10		1.76
						1.99
NM_031165.3 (Hspa8)	Heat shock protein 8	(1.6)	(2.3)	(1.1)	(1.6)	(1.2)
		2.00	2.63	1.08	1.46	1.46
		1.63	2.66	1.08	1.81	1.28
		1.15	1.47	1.06		1.09
						1.14
NM_022310.2 (Hspa5)	Heat shock 70kD protein 5 (glucose-regulated protein)	(1.6)	(1.3)	(−1.2)	(1.3)	(1.2)
		1.27	−1.15	−1.04	−1.89	2.03
		1.79	1.53	−1.92	2.48	1.02
		1.82	1.37	1.23		−1.07
						−1.17
NM_010480.3 (Hspca)	Heat shock protein 1, alpha	(1.6)	(2.1)	(−1.1)	(1.4)	(1.2)
		1.92	1.82	1.07	1.32	1.39
		1.37	2.62	−1.44	1.45	−1.28
		1.56	1.97	1.02		1.13
						1.22
NM_010477.2 (Hspd1)	Heat shock protein 1 (chaperonin)	(1.2)	(1.5)	(−1.1)	(1.3)	(1.1)
		1.23	1.36	−1.04	1.02	1.37
		1.15	1.42	−1.10	1.49	1.14
		1.14	1.57	−1.02		1.16
						1.09
NM_008303.2 (Hspe1)	Heat shock protein 1 (chaperonin 10)	(1.1)	(1.4)	(−1.1)	(1.4)	(1.1)
		1.16	1.37	−1.21	1.22	1.22
		1.11	1.39	−1.15	1.55	−1.10
		1.09	1.51	−1.00		1.31
						1.12
NM_013467.1 (Aldh1a1)	Aldehyde dehydrogenase family 1, subfamily A1	(1.3)	(1.5)	(1.1)	(1.5)	(1.5)
		−1.13	1.71	−1.03	1.33	2.29
		1.26	−1.23	1.10	1.56	−1.05
		1.64	2.01	1.21		1.55
						1.40
NM_011921.1 (Aldh1a7)	Aldehyde dehydrogenase family 1, subfamily A7	(1.3)	(1.6)	(1.1)	(1.3)	(1.4)
		1.04	1.48	−1.05	1.24	1.54
		1.48	1.33	1.25	1.39	1.36
		1.36	1.84	1.18		1.40
						1.34
NM_008898.1 (Por)	P450 (cytochrome) oxidoreductase	(3.6)	(3.4)	(3.3)	(3.6)	(4.4)
		2.45	3.57	3.23	3.76	4.84
		5.02	1.94	3.54	3.48	3.48
		3.27	4.57	3.21		5.67
						3.63
NM_011318.1 (Apcs)	Serum amyloid P-component	(1.5)	(−1.3)	(1.3)	(1.2)	(2.9)
		2.28	−1.44	1.73	−1.26	2.91
		1.05	−1.03	−1.05	1.71	5.26
		1.21	−1.30	1.21		2.20
						1.27
NM_013701.1 (Ugt1a2)	UDP glucuronosyltransferase 1 family, polypeptide A2	(1.3)	(1.6)	(1.2)	(1.4)	(2.1)
		1.24	1.58	−1.01	1.53	2.30
		1.22	1.56	1.29	1.33	2.00
		1.43	1.80	1.24		2.20
						1.72
NM_153598.1 (Ugt2b34)	UDP glucuronosyltransferase 2 family, polypeptide B34	(2.2)	(1.9)	(1.8)	(1.9)	(1.9)
		1.68	2.01	1.90	1.75	2.24
		2.44	1.27	1.74	1.98	1.48
		2.55	2.41	1.76		1.93
						1.94

Shaded genes are Keap1-Nrf-2 dependent in mouse liver (Kwak et al., 2003). Other genes in this table are also closely related to Nrf-2 dependent genes. In particular, heat shock protein 8 and UDP glucuronosyltransferase 2 family, polypeptide B34 are likely analogous to heat shock protein, 84 kDa 1 and UDP glucuronosyltransferase 2 family, member 5, respectively from Kwak et al. (2003). The values are relative to control animals and the values in parentheses are the mean of the individual values reported below the means.

mRNA Real-time PCR Study

Results of real-time PCR-determined mRNA levels of Cyp2c29, Gstmu1, and Hspa8 in CBZ-treated mouse liver were consistent with results from the microarray data (Table 3). Specifically, 24 hour, 2-dose treatment of CBZ resulted in a 12.2-fold increase in Cyp2c29, a 5.3 fold increase in Gstmu1, and a 2.5-fold increase in HSPa8 mRNA levels.

TABLE 3 Real-time PCR of Cyp2c29, Gstmu1, and Hspa8 mRNA levels in CBZ-treated mouse liver

	mRNA level (relative to GAPDH)
	8 hour		24 hour
Gene name	Control (n = 3)	CBZ-Treated (n = 4)	Control (n = 3)	CBZ-Treated (n = 4)
Cyp2c29	0.46 ± 0.054	3.30 ± 0.76	0.47 ± 0.090	5.80 ± 1.50
Gstmu1	0.67 ± 0.11	1.37 ± 0.23	0.71 ± 0.046	3.70 ± 0.75
Hspa8	0.17 ± 0.018	0.44 ± 0.066	0.27 ± 0.051	0.68 ± 0.15

*Different than control with p < 0.01 using the student test.

DISCUSSION

Treatment of mice with CBZ or PHN led to the induction of several dozen genes. Many of these genes, such as those for cytochromes P450, correspond to well-known effects of these drugs. Others may represent biomarkers of cell stress. In the literature, potential markers of oxidative stress include activation of transcription factors such as NF-κ B (Pahl and Baeuerle, 1994), AP-1 (Abate et al., 1990), Stat-3 (Andrejko et al., 1998), PPARα (Hardardottir et al., 1995), and Nrf-2 (Kwak et al., 2003). Other responses also include induction of stress proteins such as heat shock proteins (Leung and Gershwin, 1991) or of cytokines such as interleukins (Carpagnano et al., 2004). In addition, heat shock proteins are commonly regarded as a danger signal (Seguin and Uetrecht, 2003) and amyloid proteins are acute phase proteins.

A recent study applied toxicogenomics profiling to differentiate chemical reagents known to cause hepatotoxicity and oxidative stress into three subcategories: macrophage activators (MA), peroxisome proliferators (PP), and oxidative stressors/reactive metabolites (OS/RM) (McMillian et al., 2004a, 2004b, 2005). A supervised training/testing approach was used to differentiate MA, PP, and OS/RM gene transcriptional signature patterns in rat liver. Gene expression responses were selected that best separated the training set samples from all other treated and control samples. Transcription factors that distinguished the OS/RM class, such as Gst mu1 and heat shock proteins, are related to the Keap1-Nrf2-ARE signaling pathway (McMillian et al., 2004a, 2004b, 2005), which is believed to be a response to oxidative or electrophilic stress and promote cell survival (Talalay et al., 2003; Dinkova-Kostova et al., 2005; Shen et al., 2005).

In our study, 20 out of 30 of most up-regulated genes from both CBZ- and PHN-treated mice overlap with the list of genes used as the gene set used to distinguish oxidative stressors that belong to the OS/RM class (McMillian et al., 2004a 2004b) and also overlap with the list of genes involved in the Keap1-Nrf2-ARE signaling pathway (Kwak et al., 2003). This suggests that CBZ and PHN or their metabolites are responsible for an oxidative stress response that disrupts the Keap1-Nrf2 complex and serves as the common mechanism shared by both drugs. The results are also similar to a recent study of covalent binding and gene changes induced by anticonvulsants in rats (Leone et al., 2007). In view of the immune-nature of the IDRs associated with CBZ and PHN, the changes in mRNA observed are consistent with the danger hypothesis that proposes that for something to induce an immune response it must cause cell damage/stress (Seguin and Uetrecht, 2003).

Parallel in vitro studies showed that major metabolites of CBZ and PHN, 3-OH-CBZ and 4-OH-PHN, can undergo one-electron oxidation catalyzed by a variety of peroxidases to generate phenoxyl free radicals, which can abstract hydrogen atoms from biological molecules leading to redox cycling (Lu and Uetrecht, in press). Such redox cycling could be responsible for oxidative stress, as well as oxidation sulfhydryl groups on Keap1 and initiate the Keap1-Nrf2-ARE signaling pathway resulting in mRNA induction of protective enzymes such as glutathione transferase, and stress response proteins such as heat shock proteins. The finding that both CBZ- and PHN-induced similar patterns of mRNA changes associated with cell stress is consistent with the clinical observation that these two drugs produce a similar pattern of IDRs as well as cross-sensitivity and suggest that the mechanisms of their IDRs are similar.

In summary, treatment of mice with CBZ and PHN led to changes in gene expression that are consistent with the danger hypothesis and may represent biomarkers that would predict that these drugs could cause IDRs. However, it is unlikely that all drugs that cause IDRs would cause the same pattern of gene changes thus leading to one consistent set of biomarkers. It is surprising that more studies of this type have not been published; it is important to determine if there are patterns of biomarkers that would predict IDR risk and, if so, how many patterns there may be.

We are grateful to Dr. Chris Bradfield (University of Wisconsin) for the microarray analysis. J.P.U. is the recipient of the Canada Research Chair in Adverse Drug Reactions. This work was funded by grants from the Canadian Institutes of Health Research.

Current address for Wei Lu: Biotransformation, Drug Safety and Metabolism, Wyeth Pharmaceuticals, 500 Arcola Road, Collegeville, PA 19426.