Prediction of CYP3A4 enzyme activity using haplotype tag SNPs in African Americans

Abstract

The CYP3A locus encodes hepatic enzymes that metabolize many clinically used drugs. However, there is marked interindividual variability in enzyme expression and clearance of drugs metabolized by these enzymes. We utilized comparative genomics and computational prediction of transcriptional factor binding sites to evaluate regions within CYP3A that were most likely to contribute to this variation. We then used a haplotype tagging single-nucleotide polymorphisms (htSNPs) approach to evaluate the entire locus with the fewest number of maximally informative SNPs. We investigated the association between these htSNPs and in vivo CYP3A enzyme activity using a single-point IV midazolam clearance assay. We found associations between the midazolam phenotype and age, diagnosis of hypertension and one htSNP (141689) located upstream of CYP3A4. 141689 lies near the xenobiotic responsive enhancer module (XREM) regulatory region of CYP3A4. Cell-based studies show increased transcriptional activation with the minor allele at 141689, in agreement with the in vivo association study findings. This study marks the first systematic evaluation of coding and noncoding variation that may contribute to CYP3A phenotypic variability.

Introduction

The genes encoding the CYP3A subfamily are located on chromosome 7 and span 200 kilobases (kb). This gene family consists of four members, CYP3A4, CYP3A5, CYP3A7 and CYP3A43, which encode enzymes responsible for the metabolism of approximately 50% of current drugs, as well as many endogenous substances.^1,2 CYP3A4 is the major isoform in adult humans, and is expressed mainly in the liver and intestinal epithelium. It can metabolize a wide range of substrates and thus variability in its expression or activity may have significant clinical consequences. CYP3A5 is expressed in the liver, intestine³ and kidney,⁴ and has similar substrate specificity to CYP3A4. CYP3A5 has large interindividual variability in expression due to an intronic single-nucleotide polymorphism (SNP), which results in a truncated nonfunctional protein. This SNP has dramatic differences in allele frequency across populations,⁵ with the most common nonfunctional allele, CYP3A5*3, occurring at high frequency in non-African populations. CYP3A7 is the fetal CYP3A enzyme that is rapidly downregulated after birth. Although CYP3A43 is ubiquitously expressed at low levels throughout the body, its role in drug metabolism is not well understood.

These enzymes have clear interindividual differences in activity. Several studies have shown that microsomal CYP3A4 content can vary anywhere from 10- to 100-fold,^6–8 though the distribution of this variation is unimodal.⁹ This unimodal distribution suggests that simple monogenic control of CYP3A4 expression is unlikely.^10,11 In a study of healthy volunteers, the heritability of the observed enzyme variability of CYP3A4 has been estimated at 90%,¹² suggesting an important role of genetic variation at the CYP3A locus or in genes coding for the transcription factors regulating CYP3A expression. Several coding SNPs have been identified in the CYP3A locus, but they explain only a minor portion of the variation seen in levels of expression or in activity as measured by probe drugs.^13,14 However, these association studies focused on a limited number of common CYP3A SNPs, usually in the coding region or untranslated regions, and were not based on SNPs detected through a systematic and comprehensive survey of sequence variation at the CYP3A gene cluster.

Sequence variation at the CYP3A gene cluster was recently surveyed in ethnically diverse population samples. This survey aimed at uncovering sequence variation in both coding and noncoding regions that are likely to harbor sequence elements regulating CYP3A expression. The identification of such elements was achieved by a combination of comparative genomics and computational analysis. More specifically, the sequence of the entire CYP3A gene cluster was compared across five mammalian species, spanning over 300 million years of evolutionary time, which revealed many strongly conserved sequences.¹⁵ The comparative genomics approach was combined with computational prediction of clusters of liver-enriched transcription factor (LETF) binding sites within the CYP3A noncoding sequence; such clusters may signal a true regulatory region, as opposed to computationally predicted single sites, which occur at high frequency throughout the genome and are likely to be false positives. Although not all the sequence elements detected by these two approaches are expected to affect expression, combining these approaches greatly increases the probability of identifying regulatory elements; as a consequence, sequence variants within such elements may underlie some of the observed variation in CYP3A expression. By resequencing these elements in samples of African Americans, European Americans and Chinese Americans, it was shown that there are unusually large differences in the haplotype structure and allele frequencies between African-American and non-African population samples. Importantly, the amount of variation was much greater in the African-American sample compared to the samples of non-African ancestry.

This study aims to determine whether the variability in CYP3A expression and activity is due to genetic variation at the CYP3A locus. Because of greater genetic variation in African Americans compared to non-Africans, we decided to focus on African Americans as our target population. In order to capture a large amount of CYP3A sequence variation, we selected haplotype tagging SNPs (htSNPs) among those detected in the above resequencing study. These htSNPs represent the minimal set of maximally informative SNPs required to capture the haplotype diversity at the CYP3A locus in any given population. As a marker of CYP3A activity, we used a single-point plasma measurement of the ratio of the primary metabolite of midazolam (1′-OH midazolam) to midazolam. This phenotype has been used extensively as both a probe for CYP3A activity as well as specifically for CYP3A4 activity.^16–18 As in vitro studies have shown that both CYP3A4 and CYP3A5 play a role in midazolam hydroxylation, this is the ideal probe for CYP3A activity.⁹ Importantly, it was shown that this ratio is strongly correlated with CYP3A4 content (r²=0.87)¹⁹ and with midazolam clearance (r²=0.89).²⁰ Moreover, the midazolam hydroxylation ratio has been used in several other phenotype/genotype association studies for CYP3A4,^16,21–23 the results of which can be compared to the results obtained in our study. In vitro cell-based assays and liver mRNA studies were conducted to confirm the regulatory role of the htSNPs found to be significant in this study.

Results

We recruited 135 African-American outpatients to determine the effect of CYP3A variation on the midazolam metabolite ratio using the liquid chromatography/mass spectrometry/mass spectrometry (LC/MS/MS) procedure described in the Methods Section. Figure 1 shows the distribution of this phenotype, with a mean of 0.075 and a standard deviation of 0.062. We observed a 20-fold range in the midazolam phenotype, which is consistent with other studies on CYP3A4 activity.^24–26 Most of these studies found that variation in CYP3A activity ranged between 10- and 30-fold. Because of the non-normal distribution of the phenotype, the data were transformed using a natural log in all statistical analysis.

Figure 1.

Phenotypic distribution. The phenotypic distribution of the midazolam ratio is shown as the actual value and the logarithmic conversion (inset) for the in vivo genotype–phenotype association study.

The median age of the study population was 61 years, and ranged from 21 to 88 years of age. There were also a higher number of women in this study than men (86 vs 49, respectively). However, the midazolam phenotype showed no correlation with sex (P=0.24). A total of 60% of the study population had a diagnosis of hypertension upon entry into the study (as assessed by chart review, medication review and patient interview) which was significantly related (by analysis of variance (ANOVA) analysis) to the midazolam phenotype (P=0.007). In addition there was a correlation between age and midazolam phenotype (P=0.0003). Both of these variables were used as covariates in subsequent analyses. No other statistically significant correlations were found with any of the other demographic information.

In order to select htSNPs, we analyzed the data from a resequencing survey in a sample of 24 African Americans. We included all SNPs found in this survey with a minor allele frequency (MAF) greater than 10% and set the minimum correlation coefficient between SNPs (r²) to 0.8. Table 1 shows the location and sequence context of the htSNPs used in this study. The MAF cutoff was chosen to omit rare SNPs, for which there is limited power to detect a phenotypic correlation. This analysis identified 26 htSNPs, of which 19 belong to singleton bins (that is no other SNP had a correlation coefficient greater than 0.8). Because of sequence context and constaints of the genotyping platforms used, one htSNP could not be genotyped. As shown in Figure 2, the allele frequencies in the present study population and the original resequencing sample are very similar. As with allele frequencies, the linkage disequilibrium (LD) patterns between htSNPs in the original resequencing data (Figure 3a) and in the present genotyping data (Figure 3b) are very similar. In both data sets, there is little LD between htSNPs, which is expected because they were selected to have relatively low correlation coefficients in the resequencing survey. However, two pairs of htSNPs (83269 and 88726 in CYP3A43 and 280707 and 283714 CYP3A5) were more strongly correlated in our genotyping data compared to the resequencing data. This may be due to variation across samples, or to the substantial difference in sample size (24 vs 135 in the resequencing and the present genotyping studies, respectively). In addition, because the degree of African ancestry in each sample is unknown, it is possible that the differences in LD between the two are due to subtle differences in population structure.

Table 1.

Location and sequence context of htSNPs

Numbering based on NG_000004
Gene	Position transcriptional start site	Position in reference sequence	Sequence	dbSNP	Region	MAF in African Americans
CYP3A43a	31621	73432	ttaatccag[G/T]tttttttcca	rs45501292	Intron 9	0.19767
	29119	75934	atcactgac[C/T]gacctccggg	rs517284	Intron 9	0.36538
	21784	83269	acattttaa[C/G]taggtgaatt	rs6960775	Intron 7	0.48828
	20338	84715	gcttttgtc[C/T]gatcactgga	rs28669087	Intron 6	0.07197
	16327	88726	gacgaagta[C/T]aaagcacttt	rs501275	Intron 4	0.49597
	10199	94854	ccagcttaa[C/G]acataataag	rs45616432	Intron 2	0.3622
	10103	94950	aaaacaaac[C/T]ttcctgcttc	rs642761	Intron 2	0.32308
	9557	95496	agcactccc[A/T]gctcaggctc	rs523407	Intron 2	0.46262
CYP3A4b	−7206	141689	tgttgacga[G/T]atcattggta	rs2737418	Upstream	0.29528
	−4830	153798	gacagtggt[G/T]gtcaatcaaa	rs7801671	Upstream	0.2126
	15756	164751	atttatctt[G/T]ctctcttaaa	rs2687116	Intron 7	0.37209
	15980	164975	actttctgc[C/T]tctatggatt	rs2246709	Intron 7	0.32661
	16248	165243	agattgtgg[C/T]ctatcacatc	rs2687117	Intron 7	0.35385
	16616	165611	ctgctgtag[C/T]ggtgctcctt	rs4646437	Intron 7	0.26446
	20233	169228	gagtggatg[G/A]tacatggaga	rs2242480	Intron 10	0.24046
	22548	171543	tctatggag[G/A]tgtgggggag	rs6956344	Intron 11	0.35227
	23084	172079	atctaccaa[C/T]gtggaaccag	rs12721620	Intron 11	0.27273
	25375	174370	ctacccagg[G/T]ttaccttgca	rs17161886	Intron 12	0.33203
	26709	175704	cctacatgg[T/–]tgaaacccca	rs33972239	Exon 13	0.4375
CYP3A5	405	253500	ttggtagtg[G/A]gaatgatttg	rs45583844	Intron 1	0.05263
	7072	260167	tgtctttca[G/A]tatctcttcc	rs776746	Intron 3	0.25954
	14776	267871	gagcactaa[G/A]aagttcctaa	rs10264272	Exon 7	0.14552
	27612	280707	aaaacgaaa[C/T]tacatccatc	rs6976017	Intron 11	0.09398
	30619	283714	attaagcaa[C/T]agcctataag	rs10249369	Intron 12	0.14662
	31123	284218	actgggaag[G/A]gttactagac	rs45618835	Intron 12	0.05639

Abbreviations: dbSNP, single-nucleotide polymorphism database; MAF, minor allele frequency.

^aCYP3A43 is in the reverse direction in this reference sequence.

^bShaded row indicates significant SNP in univariate analysis.

Figure 2.

Allele frequency comparison. The graph shows the correlation between the allele frequency for the current in vivo genotype–phenotype association study and the previously published resequencing study. Both include only African Americans. The allele frequency for the current study shows high correlation to the previous work with a R²=0.88.

Figure 3.

Linkage disequilibrium (LD) plots of the previous resequencing study and the current midazolam study. LD plots of the previous resequencing study (a) and the current midazolam study (b) are shown. Each box within the LD plots represents the r² value for the pairwise comparisons of two single-nucleotide polymorphisms (SNPs). LD ranges from white (r²=0) to black (r²=1.0). Agreement is found between the levels of LD found between studies. Spacing of the SNPs along the locus is shown along the diagonal.

To determine if any htSNP or combination of htSNPs significantly explained the variation seen in the midazolam ratio, we used general linear model (GLM) univariate analysis. The results of this analysis are presented in Table 2. The entire model, including hypertension and age as covariates, was significant (P=0.029) with an r² of 0.474, as well as one htSNP, 141689 (P=0.011). Interestingly, when hypertension and age are removed from the analysis the model is no longer significant; however the htSNP 141689 is still significant. This SNP is located 7kb upstream of the CYP3A4 transcription start site, therefore, it stands as a plausible candidate for variation in CYP3A4 expression levels. ANOVA analysis of this SNP shows that the minor T allele is associated with a higher midazolam ratio (that is high CYP3A activity). MATCH analysis at this SNP was conducted and the results are presented in Table 3, along with the effect of the derived allele to the retention destruction and/or creation at each site. From this analysis we see that the derived allele (T allele) results in the destruction of two of the eight putative transcriptional binding sites as well as the creation of a new CCAAT displacement protein/cut repeat 3 (CDPCR3) site.

Table 2.

Results on general linear model univariate analysis

SNP	F	P-value	Gene
73432	1.015	0.37
75934	2.151	0.12
83269	2.216	0.12
84715	1.293	0.26
88726	1.543	0.22	CYP3A43
94854	2.482	0.09
94950	2.143	0.12
95496	0.612	0.55
141689	4.740	0.01
153798	2.778	0.07
164751	0.985	0.38
164975	1.115	0.33
165243	1.491	0.23
165611	0.031	0.97	CYP3A4
169228	0.273	0.76
171543	2.598	0.08
172079	0.113	0.89
174370	0.253	0.78
175704	0.487	0.62
253500	1.289	0.26
260167	0.090	0.91
267871	1.356	0.26	CYP3A5
280707	0.508	0.48
284218	1.028	0.31

Abbreviation: SNP, single-nucleotide polymorphism.

r² =0.474; P-value=0.029.

Table 3.

MATCH output of SNP 141689

Transcriptional binding factor	DNA sequence
c-Maf	taaccTGTTGacgagatca	Retained
PPARα:RXR-α	taacCTGTTgacgagatcat	Destroyed
Tax/CREB	acctgtTGACGagat	Retained
CREB	ctgtTGACGaga	Retained
CREBATF	tTGACGaga	Retained
NF-Y	gagatcATTGGtattt	Retained
Cart-1	agaTCATTggtatttata	Retained
GATA-4	AGATCattggta	Destroyed
CDPCR3	CGATAtcattggtat	Created

Abbreviations: PPARα:RXRα, peroxisome proliferator-activated receptorα:retinoid × receptorα; CDPCR3, CCAAT displacement protein/cut repeat 3; c-Maf, musculoaponeurotic fibrosarcoma oncogene homolog; CREB, cAMP response element binding protein; ATF, activating transcription factor 1; NF-Y, Nuclear Factor Y; Cart-1, cartilage homeoprotein 1.

Upper case letter denote core sequence within the site sequence.

Bold letters denote location of the SNP within the sequence.

We also tested whether variants known to affect CYP3A5 expression (CYP3A5*3, *6 and *7) as a group are correlated with hypertension or midazolam ratio. The ANOVA analysis looking at the association between hypertension and the number of functional copies of CYP3A5 was not significant.

We analyzed the intergenic sequence between CYP3A4 and CYP3A43 genes. The sequence for this genomic region was not included in the original resequencing, hence, we had no information about whether the variation in this region was tagged by htSNPs we genotyped in this study. In this region, we identified two high-probability clusters of predicted binding sites for LETFs. These two peaks were located at nucleotide positions 132736–133806 and 136085–136601 respectively (that is approximately 13 and 16 kb upstream of the CYP3A4 transcriptional start site). These clusters may reflect the presence of a sequence element regulating CYP3A expression, but they have not been surveyed for sequence variation. Therefore, we selected 16 individuals with the lowest midazolam ratio and 16 individuals with the highest midazolam ratio; these samples identify the top and bottom 12% of the phenotypic distribution. The two segments containing the clusters of predicted TF binding sites were resequenced in each of the 32 individuals to assess differences in allele frequency and haplotypes between the extremes of the phenotypic distribution. We compared the allele frequency distributions of these novel SNPs between the upper and lower tails of the phenotype distribution. No significant allele frequency differences were found. We also used PHASE (version 2.0) to infer haplotypes and determine haplotype association with phenotype for these SNPs using the case–control test to assess association. The test compares haplotype distributions between two groups to determine if any difference in distribution exists. No significant association was detected. In addition, we used MATCH to determine the location and core sequence of putative transcriptional binding sites and identified four novel SNPs in this region. Table 4 shows the position of each of these SNPs and the predicted effect on transcriptional binding sites. From this analysis we found that three sites were destroyed (two hepatic nuclear factor-3β (HNF3β) sites and one C/EBP-β site) and one site created (TATA site) by the minor alleles at these locations.

Table 4.

MATCH output of intergenic region

SNP numbera	Transcriptional binding factor	Site sequenceb,c
132798	HNF3β	caaccAAAAAaagcc	Destroyed
132798	HNF3β	aaccaAAAAAagccc	Destroyed
132798	TATA	accaTAAAAa	Created
132803	HNF3β	caaccAAAAAaagcc	Retained
132803	HNF3β	aaccaAAAAAagccc	Retained
133161	HNF3β	catatAAACAgaacc	Retained
133161	HNF3β	gaaccAAAGAcaaaa	Retained
133472	C/EBP-β	gtgTTCCTctatgt	Destroyed

Abbreviations: SNP, single-nucleotide polymorphism; HNF3β, hepatic nuclear factor-3β. C/EBP, CCAAT/enhancer binding protein β.

^aNumbered according to GenBank accession number NG_000004.

^bUpper case letters denote core sequence within the site sequence.

^cBold letters denote location of the SNP within the sequence.

To evaluate the effect of the htSNP 141689 SNP on transcription, human HepG2 and LS180 cells were transfected with the −13 kb CYP3A4-LUC wt (wild type) and variant plasmids (Figure 4). The variant 141689T allele showed higher basal activity in both the liver and intestinal cell model. Moreover, the pregnane X receptor (PXR) and rifampin-induced transcriptional activities (PXR and rifampin alone and in combination) of the 141689T allele were significantly higher than the 141689G allele in both the LS180 and HepG2 cells.

Figure 4.

CYP3A4 transcriptional activity of haplotype tagging single-nucleotide polymorphisms (htSNP) 141689 in HepG2 and LS180 cell lines. HepG2 cells (a) and LS180 cells (b) were transfected with the −13 kb CYP3A4-LUC wt and variant allele plasmids or the empty vector plasmid (pGL3basic) with (+) or without (−) cotransfected hPXR (pregnane × receptor) and treated with dimethylsulphoxide (DMSO) vehicle (−) or 10 µM rifampin (+) and luciferase activity was normalized to total protein. The mean and standard deviation in HepG2 cells, and the mean and range in LS180 from a representative experiment are shown. * indicates P≤0.05, and ** indicates P≤0.06 when comparing activity of the major and minor allele using the t-test: (assuming equal variances).

African-American donor livers were genotyped to determine the effect of htSNP 141689 on hepatic CYP3A4 mRNA levels. Among male African-American livers (n=32), those heterozygous at 141689 showed somewhat higher levels of CYP3A4 mRNA compared to the homozygous 141689 TT samples (Figure 5a). However, the 141689 TT homozygote livers had a significantly lower CYP3A4 mRNA levels compared to livers with at least one 141689G allele. Similar results were seen when the analysis was performed in the combined group of male and female African-American donor livers (n= 48) (Figure 5b). These findings contradict our in vivo phenotype–genotype association results and the cell-based transcriptional activation results. This may be due to the unknown drug and disease status of some of the liver donors used in this study.

Figure 5.

Association of the 141689G/T single-nucleotide polymorphism (SNP) with CYP3A4 mRNA levels in human liver. The results are shown as box plots where the box represents the middle 50% of the data and the whiskers represents the spread of the remaining data. The line in the center represents the median. P-values for differences between groupings of the major vs minor alleles were determined using the t-test, Wilcoxon’s test or the Kruskal–Wallis test. (a) Relative CYP3A4 mRNA levels (log₂ values) were determined in 32 male African-American donor livers and plotted against the haplotype tagging SNP (htSNP) 141689 genotypes. (b) Relative CYP3A4 mRNA levels were determined in 48 African-American donor livers (including both men and women) and plotted against the htSNP 141689 genotypes.

Discussion

Studies of CYP3A haplotypes have been limited to genotyping of functional alleles,²⁷ or genotyping SNPs interspersed throughout the locus.²⁸ By using a htSNPs approach based on a full resequencing study of conserved sequences across distantly related species, the present study represents the first systematic and comprehensive test of the contribution of genetic variation within the CYP3A gene cluster to variation in CYP3A enzyme function. In this study we found a significant association between the midazolam ratio phenotype with htSNP 141689 found −7206 bp upstream of CYP3A4. Further cell-based in vitro studies showed that the 141689T allele was associated with increased basal and PXR inducible CYP3A4 transcriptional activity, suggesting a role for this SNP in basal and inducible CYP3A4 hepatic expression.

Because of the significant association, we used MATCH analysis to infer the role of htSNP 1415689 in transcription factor binding (results shown in Table 3). The most notable of these is the destruction of the peroxisome proliferators-activated receptor (PPAR)α/retinoid × receptor (RXR)α and GATA-4 sites and the creation of the CDPCR3 site. Induction of CYP3A4 by endogenous compounds such as steroidal hormones or xenobiotics is known to be mediated by constitutive androstane receptor (CAR),^29,30 PXR^31,32 and the vitamin D receptor.³³ Constitutive expression of CYP3A4 may also be regulated through these nuclear receptors.³⁴ Both PXR and CAR require interaction with RXRα to regulate transcription. Animal studies have shown that mice deficient in RXRα have reduced basal expression of CYP3A4.³⁵ PPARα is known to form a heterodimer with RXRα and regulates CYP4A expression in mice.³⁶ However, this regulatory mechanism was not seen in human hepatocyte cultures.³⁷ Further studies are needed to determine the role of these TFB sites and cognate binding factors to CYP3A4 regulation. It should be noted that htSNP 141689 (−7206) is located just downstream of the XREM regulatory region (−7836 to −7607 bp upstream of the transcriptional start site) of CYP3A4. This region, in conjunction with a proximal XREM directs PXR and CAR-mediated transcription of CYP3A4.³⁸ Currently the role of the PPARα/RXRα heterodimer, GATA-4 and CDPCR3 are not known in relation to CYP3A regulation.

Although the CYP3A4 5′-flanking region is more than 35.8kb in length, most regulatory studies have focused on the proximal 12kb sequence immediately upstream of the gene. In a study by Matsumura et al.³⁹ focused on the region 12kb upstream of CYP3A4, the investigators identified the constitutive liver enhancer module of CYP3A4 (CLEM4) located 11.5 kb upstream of the CYP3A4 gene. However, in our study we resequenced regions between CYP3A4 and CYP3A43, which has not been previously investigated. MATCH analysis of the four novel SNPs found in this region revealed that these SNPs are predicted by in silico analysis to affect putative transcription factor binding sites for HNF3β and C/EBP-β, and one SNP leads to the creation of a TATA site. The HNF-3 proteins are involved in the regulation of numerous liver-specific genes.⁴⁰ An HNF3β site has previously been identified −195 to −186bp upstream of CYP3A4 and disruption of this site caused downregulation of basal CYP3A4 expression in transient transfection assays.⁴¹ As the HNF3β sites found in our analysis are further upstream than those in previous work, further in vitro studies are needed to determine the role of these regions in regulation of the CYP3A locus.

In this study we included SNPs within CYP3A5 that are known to cause a loss of function. The most highly studied, CYP3A5*3 within intron 3 of the gene, creates a cryptic slice site resulting in a nonfunctional protein. This nonfunctional variant is at high frequency in Caucasians, with approximately 10–30% of individuals in this population expressing CYP3A5, whereas approximately 60% of African American express CYP3A5.^5,42 In addition, two other loss of function variants, CYP3A5*6 and CYP3A5*7, are found solely in populations of African ancestry.⁴³ Although none of these SNPs alone or in combination with other htSNPs were shown to contribute significantly to phenotype variation in our study, the effect of these SNPs on drug metabolism is not entirely clear. Although the CYP3A5*3 variant has shown clear association with midazolam metabolism in vitro,^5,9,44 the in vivo evidence is contradictory, with a bulk of the studies showing no association of midazolam metabolism with CYP3A5 genotype.^17,45–48 Clearly the role of both CYP3A4 and CYP3A5 in midazolam metabolism is complex and may involve variation in the regulation of these genes,⁹ or difference in CYP3A4 content related to CYP3A5 genotype.⁴⁹

During our initial analysis, we found an association of hypertension to the midazolam metabolite ratio. This finding was interesting given the numerous studies looking at the role of CYP3A5 and hypertension. Significant associations have been found in studies looking directly at the association of the hypertension phenotype in African Americans to the known functional SNPs in CYP3A5.^50–52 A previous publication from our laboratory showed that, like other variants influencing risk to hypertension,⁵³CYP3A5*1 has large population difference in allele frequency and has a strong latitudinal cline with populations close to the equator having moderately high frequency of the allele whereas populations at latitudes further from the equator have low allele frequencies.⁵⁴ It was postulated that this may be due to a selective advantage for salt retention in hot climates that have scarce water supplies, which may contribute to salt-sensitive hypertension in populations of African ancestry. In contrast with these findings, our data did not show a significant association between the number of functional copies of CYP3A5 and either hypertension or the midazolam phenotype. This negative result may simply be due to insufficient power, or the inability to classify the type of hypertension within our study population. A systematic evaluation of salt-sensitive hypertension in African-Americans and CYP3A variants is needed to definitively answer this question.

Both the in vivo clinical genotype/phenotype study and the in vitro transcriptional cell-based assay show an association of the htSNP 141689 with increased CYP3A4 enzyme activity. However the in vitro liver mRNA study showed contradictory results. This may be due to several factors, namely the unknown drug and disease status of the liver donors. In the clinical genotype/phenotype study a thorough evaluation of each subject’s drug and disease status was performed before inclusion. If a subject had a disease which could affect drug clearance and/or enzyme function (such as liver or renal dysfunction) they were excluded from the study. In addition, any subject on drugs which induced or inhibited CYP3A was excluded. In this way the clinical phenotype/genotype study allowed more conclusive evaluation of the CYP3A phenotype given the strict inclusion criterion.

If, indeed, this SNP lies within a regulatory element responsible for constitutive activity of CYP3A4, this element may have alternate function in an induced or inhibited environment. The cell types used may also explain these results. Maruyama et al.⁵⁵ showed differential mRNA expression of CYP3A4 in HepG cells compared to human fetal liver cells in induced states, with the human fetal liver cell showing increase in mRNA to the known CYP3A4 inducer rifampin and no change seen in the fetal liver cells. In addition, degradation of enzyme due to harvesting and cold storage may also play a factor in these results.⁴⁶

There are a few limitations to the current study. First, a replication cohort would provide definitive answers to the involvement of the SNP at position 141689 in the regulation of the CYP3A locus. However, such replication studies are fiscally prohibitive and few investigators conduct studies in exclusively African-American populations. This study reflects the first association study based on population genetics and comparative genomics data from a systematic survey of CYP3A sequence variation.¹⁵ The hope is that our study will serve as hypothesis generation for further investigations into the role on noncoding variation at this locus. Secondly, the contradictory findings of the in vitro liver study also limit the definitive association of this SNP with the regulation of the CYP3A locus. However, as stated previously, the unknown drug and disease status of the liver donors as well as the experimental conditions may make this a less robust assessment of phenotype. Though inconsistencies between in vitro and in vivo findings are not uncommon in the association literature, further studies are needed to determine if the SNP at position 141689 binds the predicted transcriptional factors. Lastly, although it is possible that the association we found through the GLM was a false positive given the number of statistical tests, the confirmation of association by the cell-based assays make this less of a concern.

In conclusion, this study has provided a systematic and comprehensive investigation into the role of genetic variation in the CYP3A locus to hepatic CYP3A4 activity. We found that htSNP 141689, found −7206 upstream of the CYP3A4 gene, significantly affects the transcriptional activation and enzyme activity of CYP3A4. Further studies are needed to determine the effect of this novel CYP3A4 promoter SNP to CYP3A4 basal and inducible expression in vitro and in vivo. In addition, as the novel SNPs we identified in the CYP3A locus can still not explain the full extent of CYP3A4 human variation, our findings strengthen support for the hypothesis that SNPs in transcription factors regulating CYP3A4 are likely to be important contributors to human variation in CYP3A4 activity.

Methods

Patient population

A total of 135 African-American subjects were recruited through the Gastrointestinal Procedure Unit at the University of Chicago Hospitals, where a single intravenous dose of midazolam is commonly utilized for conscious sedation. All subjects were screened according to medication use, medical history and clinical laboratory evaluation by both chart review and patient interview. Inclusion criteria included: a single intravenous bolus dose of 2–3 mg of midazolam given within 5 min for an elective procedure, age greater than 18 years, no known concomitant medications that would inhibit or enhance the CYP3A enzymes and consent before procedure. Exclusion criteria included: multiple doses of midazolam, diagnosis of abnormal liver or renal function, abnormal renal or hepatic laboratory values (BUN, serum creatinine, liver-specific enzymes), smoking or heavy drinking, and use of interfering medications. Every subject gave written consent before dosing of midazolam and the Institutional Review Board of the University of Chicago approved this study. Two 10 ml blood samples were taken from each subject by venipuncture 30min after midazolam dosing. Blood samples were placed in purple top Vacutainer tubes, containing ethylenediaminetetraacetic acid, and one 10 ml sample of whole blood was frozen at −20°C for DNA extraction. The other 10 ml sample was centrifuged and the plasma was frozen at −20°C until time of analysis.

Genotyping procedure

Different genotyping methods were used depending on the type and sequence context of each SNP. A published restriction fragment length polymorphism assay was used to genotype the CYP3A5*3 polymorphism.⁵⁴ A total of 19 SNPs were genotyped at DNAprint Genomics (http://www.dnaprint.com/), using the Orchid SNP Stream-Ultra-High Throughput (UHT) platform (primers in Supplementary Table 1). This method could not be optimized for six SNPs; therefore these SNPs were genotyped by either a TaqMan assay or a single base extension (SBE) assay. Because of the sequence similarities between members of this gene family and the stringent product lengths needed for these assays, all assays were performed through nested PCR. All PCR primers were designed based on the GenBank accession number NG_000004; all nucleotide positions in this publication are numbered according to this sequence.

The optimal TaqMan primer/probe set (Supplementary Table 2) was identified using the Primer Express 2.0 software (Applied Biosystems, Foster City, CA, USA). The TaqMan assay was carried out in a 25 µl reaction containing 1 µl of PCR Product, 1 × Universal Master Mix (Applied Biosystems), 10µm of each primer and 2µm of each fluorescent probe. The reaction was run on a MJ Research thermocycler at 50 °C for 2 min, 95 °C for 10 min, followed by 35 cycles of 92 °C for 15 s and 60 °C for 1 min. Genotype analysis was performed on the ABI SDS 7700 PCR machine with ABI PRISM 7700 software, using a standard procedure of automatic allele discrimination.

SBE assays involved PCR amplification (primer in Supplementary Table 3) of genomic DNA followed by enzymatic cleanup with shrimp alkaline phosphatase and exonuclease I (ExoI). The final SBE reaction was carried out in a 12 µl reaction containing 1 × Sequenase reaction buffer (GE Healthcare Bio-Science Corp., Piscataway NJ, USA), 7.2 µl of purified PCR product, 250 µm of each ddNTP (GE Healthcare Bio-Science Corp.), 1 µm allele-specific oligonucleotide, and 2.5 U of Sequenase (GE Healthcare Bio-Science Corp.). The reaction was run on a MJ Research thermocycler at 95 °C for 2 min, followed by 60 cycles of 96 °C for 30 s, 55 °C for 30 s, and 60 °C for 30 s. Genotypes were determined by denaturing high-performance liquid chromatography (dHPLC), which involved injection of each reaction into a preheated reversed phase column (DNA-Sep; Transgenomic, San Jose, CA, USA) on a Wave 3500HT DHPLC System (Transgenomic Inc., Omaha, NE, USA). DNA was eluted on a linear acetonitrile gradient consisting of buffer A (0.1 moll⁻¹ triethylammonium acetate, TEAA)/buffer B (0.1 moll⁻¹ TEAA, 25% acetonitrile). Data acquisition was controlled by Navigator software (version 1.6.2; Transgenomic Inc.).

Phenotyping procedure

The concentrations of midazolam and its major metabolite, 1-hydroxy (1-OH) midazolam, were measured in plasma using a published liquid chromatography/mass spectrometry (LC/MS) method with some modifications.⁵⁶ Both patient samples and concentration standards were analyzed using the following method. Briefly, a 1 ml aliquot of plasma was incubated with 2000 U ml⁻¹ β-glucuronidase to cleave the glucuronide-conjugated metabolite. Using this method we measure the concentration of both the 1-OH midazolam and the 1-OH midazolam glucuronide. Samples were then spiked with 10µl of prazapam (100 mg ml⁻¹ dissolved in ethanol) and extracted as previously described by Frerichs et al.⁵⁶ For each sample and standard, 100 µl was injected into the detection system for analysis. Concentration plasma standards were made by adding appropriate amounts of both midazolam and 1-OH midazolam to blank human plasma at concentrations ranging from 150 to 1 ng ml⁻¹. All samples were measured in duplicate with less that CV > 0.15 between samples.

The concentrations of all samples and standards were determined by an Agilent 1100 series HPLC coupled to an ABI 2000 triple quadrapole LC/MS fitted with an electro-spray ionization interface. The samples were first separated using a Luna C₁₈ reverse-phase chromatography column (3 µm particle size, 100 × 46 mm i.d.; Phenomenex, Torrance, CA, USA) with an isocratic mobile phase of acetonitrile:ammonium acetate (50:50, pH 4.7) at 1 ml min⁻¹. The MS conditions were optimized using a direct infusion of standard solutions prepared in mobile phase. LC/MS quantitation was performed at the following conditions: source temperature 450 °C, curtain gas 40ml min⁻¹, collision gas 6 ml min⁻¹, nebulizer voltage 6 V, ion source 1 gas 50 ml min⁻¹, ion source 2 gas 50 ml min⁻¹, declustering potential 51 V, focusing potential 370 V, entrance potential 11 V, collision energy 29 V and collision cell exit potential 8 V. Nitrogen and argon were used as the nebulizer and collision gases, respectively. Cone voltage and collision energies were optimized for each analyte by full scan acquisitions. All analyses were performed in positive ion mode. Data acquisition and quantitative analysis were performed by Analyst software (version 1.4.1; Applied Biosystems).

Resequencing of the CYP3A43-CYP3A4 intergenic region

Analysis of the intergenic region between the CYP3A43 and CYP3A4 genes revealed new clusters of predicted transcriptional factor binding sites (Figure 6). The two regions closest to CYP3A4, located 13 and 16 kb upstream of the gene, were resequenced in the upper and lower ends of the phenotypic distribution (total of 32 individuals) to assess genomic differences between these two groups. PCR products were cleaned using SAP and ExoI. Dye terminator sequencing was performed with the ABI Big Dye Terminator Cycle sequencing kit and the products were analyzed with an ABI 3730 automated sequencer (Applied Biosystems). The Phred-Phrap-Consed package (version 5.03) was used to assemble and analyze the sequences.

Figure 6.

Predicted TFB sites in the CYP3A intergenic region. Graphical analysis by Cister shows the location of TFB sites between CYP3A4 and CYP3A43. The peaks represent TFB site cluster probability with each TFB site shown as a designated colored line within each peak. The location of each peak is marked as distance upstream from CYP3A4 transcriptional start site.

Statistical analysis

We identified LD patterns and picked htSNPs using the program LDSelect,⁵⁷ which uses r² (a measure of LD between pairs of SNPs), to group SNPs that are highly correlated with each other. This method reduces the number of SNPs that need to be typed while still capturing the majority of the variation present in the population. Figure 7 shows the SNPs that were binned together and the htSNP chosen to tag each bin across the CYP3A locus. Seven haplotype tagging bins were identified, each spanning only one of the CYP3A genes. In addition, 19 singleton bins were found (shown as red dots); singleton bins contain SNPs that are not highly correlated to any other SNPs in the locus, but which may still explain phenotypic variation alone or in combination with other htSNPs.

Figure 7.

Haplotype tagging single-nucleotide polymorphisms (htSNP) binning by LDSelect across the CYP3A locus. The arrows indicate the direction of each CYP3A gene along the chromosome 7. Each colored bar represents the location of the haplotype tagging bin with the SNPs found within each bin shown in the corresponding color and the htSNP for each bin designated by the bolded SNP number. Superscript letters represent the following: A, CYP3A4*1B; B, CYP3A5*3 and C, CYP3A5*7. Seven haplotype tagging bins were identified along with 19 singleton bins (shown as red dots).

We initially assessed the correlation of each demographic factor with the midazolam phenotype using ANOVA and simple regression. The demographic factors that showed a significant correlation were then used as covariates in the subsequent analyses. We then performed GLM univariate analysis to determine which htSNP best predicted the midazolam phenotype.

Because of the previously reported association between CYP3A5 variants and hypertension,^52,58 we also looked at the correlation between hypertension and midazolam phenotype, respectively, with the nonfunctional CYP3A5 alleles, CYP3A5*3, *6 and *7, as a group. We used PHASE to infer the haplotype phase in our sample.⁵⁹ We then determined the number of functional copies of CYP3A5 for each individual. For example, if two nonfunctional alleles were found on the same chromosome, as determined by PHASE, and only the CYP3A*1 allele was found on the second chromosome, this individual would be scored as having one functional copy of CYP3A5. If, however, an individual was homozygous for any of the nonfunctional alleles or if different nonfunctional alleles were found on the two chromosomes, the individual was scored as having no functional copies of CYP3A5. Of the 270 chromosomes investigated in this study, only 3 chromosomes were found that contained two nonfunctional alleles of the same chromosome. A total of 23 individuals were heterozygous for two of these functional SNPs with the nonfunctional SNPs occurring on different chromosomes. We then used ANOVA and χ²-analysis to look at association of the number of functional copies to the midazolam ratio and hypertension. All analyses were performed using the SPSS software package (version 15.0.0). A P-value of less than 0.05 was deemed significant, and there was no correction for the multiplicity of statistical testing.

As the human liver mRNA levels were not normally distributed, group differences were analyzed nonparametrically using the Wilcoxon’s rank sum test (W) to compare binary groups (for example, GG + GT vs TT). The Kruskal–Wallis (KW) test was used to compare the three genotypes for each polymorphism (for example, GG vs GT vs TT). All statistical calculations were performed using statistical program R: A Language and Environment for Statistical Analysis (http://www.R-project.org).

MATCH analysis

The MATCH version 11.1 tool (TRANSFAC Professional suite, version 11.1) was used to search for potential TFB sites in the 50 bp region surrounding htSNP 141689. MATCH uses a library of nucleotide distribution matrices of aligned TF binding sequences that are obtained by in vitro selection studies to predict TF binding sites in this sequence. Our search was restricted to the use of high-quality matrices only (that is to exclude highly abundant matrices). We used the vertebrate nonredundant profile with the minSUM cutoff function. The minSUM cutoff minimizes both false positive and false negative matches. Cutoffs are based on matrix similarity scores which describe the quality of a match between a matrix and a part of the input sequence. Analogously, the core similarity denotes the quality of a match between the core sequence of a matrix (that is the five most conserved positions within a matrix) and a part of the input sequence. The core similarity and matrix similarity scores for the predicted TF binding sites are calculated using the MATCH algorithm described by Kel et al.⁶⁰

Plasmids

The −13 kbCYP3A4-LUC (wt) plasmid containing the entire CYP3A4 promoter, including the proximal and distal PXR response elements, inserted in pGL3 Basic (Promega, Madison, WI, USA) was kindly provided by Dr Christopher Liddle.³⁸ This plasmid contained the wild-type 141689G allele. The CYP3A4 141689T variant (var) allele was created directly from −13 kbCYP3A4-Luc (wt) by site-directed mutagenesis (QuickChange; Stratagene, La Jolla, CA, USA) using a variant primer (sense primer, 5′-CCATTCTCCTTTAACCTG TTGACGATATCATTGGTATTTATAC-3′, htSNP 141689 bolded). The ∓13 kbCYP3A4-Luc var plasmid was sequenced to confirm base changes.

Transient transfection studies

Human hepatoblastoma cells, HepG2 (American Type Culture Collection), were grown in minimal essential medium (Alpha-MEM; Biowhittaker, Waltersville, MO, USA) containing 10% fetal bovine serum, penicillin and streptomycin. On day 1, HepG2 cells were seeded into 24-well plates at a density of 0.6 × 10⁶ cells per well. After 24 h, cells were transfected with 200 ng per well of the wild-type or variant −13kb CYP3A4-Luc reporter plasmid by calcium phosphate method. LS180 colon carcinoma cells were seeded into 24-well plates at a density of 0.5 × 10⁶ cells per well on day 1. They were then transfected on day 2 by dissolving plasmid in OptiMEM (Life Technologies, Carlsbad, CA, USA) and transfecting using GenJet DNA in vitro transfection reagent (SignaGen, Gaithersburg, MD, USA). After 16 h, the cells were washed with OptiMEM then incubated for 24 h. Luciferase activities were measured with the Luciferase Reporter Assay Kit (Promega) on an OPTO-COMP I luminometer and normalized to total cell protein. The experiments were repeated in triplicate. Differences in transcriptional activation were assessed by a Student’s t-test.

Human livers

Institutional review boards and clinical research advisory committees at St. Jude Children’s Research Hospital and the University of Pittsburgh approved the use of tissue samples from organ donors. Human liver tissue was processed through Dr Relling’s laboratory at St. Jude Children’s Research Hospital and was provided by the Liver Tissue Procurement and Distribution System (NIH Contract no. N01-DK-9-2310) and by the Cooperative Human Tissue Network. Total RNA was isolated from the liver tissue from organ donors (48 African-American livers (men, n=32, women, n=16) using Trizol (Invitrogen, Carlsbad, CA, USA). First, strand cDNA was prepared from 3 µg total RNA using oligo dT primers and the Invitrogen Superscript II kit. Before real-time PCR, 20 µl of cDNA was diluted to 50 µl with DEPC-treated water.

Relative quantitation of CYP3A4 mRNA by the standard curve method

An ABI gene expression assay (Hs00430021_m1) was used for real-time PCR (rtPCR) quantitation of CYP3A4 mRNA levels. Human PPIA (cyclophilin A, VIC/MGB Probe, Primer Limited) was used to normalize the relative CYP3A4 mRNA expression according to the manufacturer’s instructions. cDNA (2 µl) from each sample was analyzed in duplicate by rtPCR on an ABI PRISM 7900HT Sequence Detection System (PE Applied Biosystems, Foster City, CA, USA). The reaction was run using 1 µl of the 20 × Gene Expression Assay mix along with 7 µl of DEPC-treated water and 10 µl of 2 × Taqman Universal PCR Master mix (with Amperase UNG; TaqMan Universal PCR Master Mix) in 20 µl final volume. Standard amplification conditions consisting of 2 min (UNG activation) at 50 °C, 10 min at 95 °C followed by 40 cycles of: 15 s (denaturing) at 95 °C and 1 min (annealing/extension) at 60 °C. Standard curves (in triplicate) were prepared for each gene using cDNA from a high expression sample serially diluted over a 3125-fold range and the amounts of CYP3A4 and cyclophilin mRNA determined by interpolation. The average CYP3A4 amount was divided by the average cyclophilin amount to obtain a normalized CYP3A4 value. One of the experimental samples was then chosen as the calibrator, and each of the normalized CYP3A4 values was divided by the normalized calibrator value to get the relative CYP3A4 mRNA expression levels for the study samples.