Skip to main content
NCBI home page
Proceedings of the National Academy of Sciences of the United States of America logoLink to Proceedings of the National Academy of Sciences of the United States of America
. 2007 Oct 31;104(45):17735–17740. doi: 10.1073/pnas.0700724104

In silico and in vitro pharmacogenetic analysis in mice

Yingying Guo †, Peng Lu †, Erin Farrell †, Xun Zhang †, Paul Weller ‡, Mario Monshouwer ‡, Jianmei Wang †, Guochun Liao †, Zhaomei Zhang †, Steven Hu †, John Allard †, Steve Shafer §, Jonathan Usuka †, Gary Peltz †,
PMCID: PMC2077071  PMID: 17978195

Abstract

Combining the experimental efficiency of a murine hepatic in vitro drug biotransformation system with in silico genetic analysis produces a model system that can rapidly analyze interindividual differences in drug metabolism. This model system was tested by using two clinically important drugs, testosterone and irinotecan, whose metabolism was previously well characterized. The metabolites produced after these drugs were incubated with hepatic in vitro biotransformation systems prepared from the 15 inbred mouse strains were measured. Strain-specific differences in the rate of 16α-hydroxytestosterone generation and irinotecan glucuronidation correlated with the pattern of genetic variation within Cyp2b9 and Ugt1a loci, respectively. These computational predictions were experimentally confirmed using expressed recombinant enzymes. The genetic changes affecting irinotecan metabolism in mice mirrored those in humans that are known to affect the pharmacokinetics and incidence of adverse responses to this medication.

Keywords: drug metabolism

A quantitative simulation demonstrated how utilization of pharmacogenomic information to individualize drug dosage has the potential to significantly improve treatment outcome and reduce the rate of attrition of drugs in clinical development (1). However, we often do not know the genetic factors responsible for interindividual variability in drug response. Typically, drug doses are adjusted empirically based upon therapeutic response or toxic effect, indicating that the initial dose was either subtherapeutic or potentially toxic. Despite its potential, clinical utilization of pharmacogenomic information is limited by our poor understanding of the genetic variables that govern variability in response (1). To enable the routine use of pharmacogenomic testing in clinical practice, efficient strategies for identifying these genetic variables must be developed.

Toward this end, we have recently demonstrated that genetic factors affecting the metabolism (2) or the pharmacodynamic response (3, 4) for clinically important drugs can be rapidly identified in mice by computational haplotype-based genetic analysis (5–7). This approach requires administering a test drug to multiple inbred strains and then measuring individual metabolites in plasma at multiple time points after dosing (2). Most drugs are metabolized by multiple pathways, and individual steps in each pathway may be catalyzed by distinct enzymes. Therefore, analysis of the rate of formation of individual metabolites in plasma across multiple inbred murine strains reduced the complexity of this biological process, which enabled the genetic factors that contribute to the overall variability in drug metabolism to be identified (2). Reduction of biological complexity increases the chance of discovering the effect that a genetic difference within a single region has on a measured phenotype (6). This enables the pattern of genetic variation within discrete regions of the mouse genome to be correlated to patterns of phenotypic variation. Thus, analysis of less complex phenotypes and discrete genomic regions makes use of the genetic relatedness among inbred strains and minimizes the confounding effect of linkage among regions in the mouse genome (8–10).

The in vivo pharmacogenetic analysis method, although powerful, is quite labor-intensive. We wanted to determine whether the genetic differences affecting drug metabolism could be more efficiently and rapidly identified by using an in vitro analysis system that is amenable to high-throughput testing. This could be accomplished by measuring the rate of metabolite formation after in vitro incubation of medications with liver extracts prepared from 15 different inbred mouse strains. If successful, the first two steps (referred to as phase I and phase II reactions) in drug metabolism could be rapidly analyzed.

To test this method, we selected two drugs, testosterone and irinotecan, whose metabolism has been previously characterized. However, the information about their metabolism was generated using existing tools, which required either a very prolonged effort [quantitative trait locus analysis in mice (11, 12) for testosterone] or the Herculean task of cloning, expressing, and testing different allelic forms of 16 different UDP-glucuronosyl transferase (UGT) enzymes [irinotecan (13)]. The availability of whole-genome sequence for mammalian organisms has further increased the workload required to analyze individual genes within enzyme families. The large number of mammalian UGT (31 human and 23 murine) (14) and Cyp450 [92 murine genes within National Center for Biotechnology Information (NCBI) build 36] enzymes makes it prohibitively difficult to test the ability of each expressed recombinant enzyme, and subsequently allelic variants, to mediate a particular biotransformation reaction for a panel of drugs.

We selected testosterone for analysis, because cytochrome P450 enzymes, which are located in the endoplasmic reticulum (i.e., phase I reaction), are known to catalyze the rate-limiting steps in its metabolism. We selected irinotecan, a chemotherapeutic agent used for treatment of colorectal cancer and glioma (reviewed in ref. 15), because interindividual differences in its pharmacokinetics and in the incidence of adverse responses have been shown to be due to genetically determined differences in the rate of glucuronidation of this drug by cytosolic UGT1A1 enzymes (i.e., phase II reaction) (13). Irinotecan is converted to its active metabolite [7-ethyl-10-hydroxy-camptothecin (SN-38)], which is a topoisomerase I inhibitor that causes cancer cell death by inducing DNA strand breaks, and this metabolite is inactivated by glucuronidation. The two main dose-limiting and life-threatening toxicities associated with irinotecan therapy are myelosuppression and diarrhea. Specifically, an allelic variant with an extra TA repeat within the TATAA element of the 5′ promoter region (UGT1A1 *28) of UGT1A1 (16) is associated with reduced glucuronidation and increased irinotecan toxicity (17, 18).

Here, we demonstrate that a murine in vitro hepatic drug biotransformation system can be used in conjunction with haplotype-based computational analysis to efficiently identify genetic factors effecting phase I and phase II drug metabolism in mice. This in vitro and in silico analysis method rapidly identified the basis for genetic variability in irinotecan metabolism in mice, which mirrored human genetic variability that contributes to interindividual differences in response to this medication.

Results and Discussion

Computational Genetic Analysis of Testosterone Metabolites Produced in Vitro.

We selected testosterone, an endogenous hormone and a therapeutic agent (19), for analysis, because it is primarily metabolized by hepatic Cyp450 enzymes. This provided an opportunity to test the ability of the in vitro/in silico system to analyze the metabolism of drugs with extensive phase I metabolic profiles. To develop the in vitro hepatic drug biotransformation system, liver microsomes were prepared from 15 inbred mouse strains. The rate of formation of hydroxylated testosterone metabolites produced after incubation of (20 μM) testosterone with the liver microsomal extracts was quantitatively assayed by liquid chromatography tandem MS (LC/MS/MS) analysis (Fig. 1A). Consistent with previous studies in mice (20), testosterone was biotransformed into seven different hydroxylated metabolites (2α-, 2β-, 6β-, 15α-, 15β-, 16α-, and 16β-). Each metabolite had a unique rate of formation in extracts prepared from the inbred strains, and these results were reproducible in three independently performed experiments [supporting information (SI) Table 2]. Because three independent measurements were performed, we could assess the significance of the differences in the rate of formation of each of the seven identified metabolites (SI Table 2). The rate of 15α- and 16α-hydroxytestosterone (OHT) formation was most variable across the inbred strains analyzed [P < 0.0001 and the coefficient of determination (R2) > 0.85 by ANOVA analysis]. However, the rate of 15α-OHT formation varied continuously across the strains, and computational genetic analysis of these data did not yield any significantly correlated haplotype blocks (P value < 0.0001) (data not shown). In contrast, the rate of 16α-OHT formation was significantly reduced in two inbred strains (LP/J and 129) relative to the other 13 strains (Fig. 1B). This metabolite was selected for further computational genetic analysis.

Fig. 1.

Fig. 1.

Open in a new tab

In vitro testosterone biotransformation. (A) Chromatograms of testosterone metabolites produced after 8 min of incubation of 20 μM testosterone with (0.2 mg of protein per milliliter) microsomes prepared from the C57B6 and LPJ strains in vitro. The peaks representing six identified hydroxylated testosterone metabolites and one unidentified metabolite (M1) are indicated. Of note, the microsomes from the LPJ strain produce a significantly smaller amount of 16α-OHT. (B and C) The amount (micromolar per milligram of protein) of 16α-OHT formed after incubation of (20 μM) testosterone with liver microsomes prepared from the indicated 15 inbred mouse strains is plotted as a function of time (B) or after 8 min of incubation (C). After incubation for the indicated time period, the amount of 16α-OHT formed was quantitated by LC/MS/MS analysis. Each data point represents the average ± SD of results obtained from analysis of three individual incubation reactions. (D) The data in C were log-transformed and analyzed by the haplotype-based computational genetic method, and the output is shown. A representative six of the 23 haplotype blocks having the highest correlation (P = 3.2 × 10−5) with the rate of metabolite formation, along with all other blocks with P < 0.0001, are shown. A complete list of the haplotype blocks with P = 3.2 × 10−5 is shown in SI Table 3. For each predicted block, the chromosomal location, number of SNPs within a block, its gene symbol, and an indicator of whether its mRNA is expressed in liver are shown. The haplotype for each strain is represented by a colored block, and the blocks are presented in the same order as the phenotypic data. For each block, the calculated P value measures the probability that the phenotypic data would have the same degree of association with strain groupings when the mean trait values are the same for each group. The measured level of mRNA expression of the indicated gene using microarrays in liver tissue obtained from the strain with the highest measured value is shown. Each number is the average of three independent measurements for the strain with the highest level of expression, and liver tissue from 10 inbred strains was analyzed.

16α-OHT is a low-abundance metabolite that accounts for 0.5–3% of the metabolites produced in vitro (SI Table 2). Haplotype-based computational genetic analysis (5–7) was used to identify genetic factors contributing to the strain-specific differences in the rate of 16α-OHT production. Twenty-three haplotype blocks were identified where the two strains (LP/J and 129/SvJ) with the lowest rate of 16α-OHT production shared a unique haplotype that was distinct from the other 13 strains (Fig. 1D and SI Table 3). These blocks have the lowest calculated P value of 3.2 × 10−5 for this data set, but all blocks with a P value <0.0001 were evaluated. Because 5,225 blocks were evaluated, a threshold of P value <0.0001 limits the number of false-positive predictions yet provides a reasonable number of candidate genes for evaluation. However, other than the 23 blocks mentioned above, no other blocks were found in which the strains with a low or a high rate of 16α-OHT production shared distinct haplotypes, regardless of how the strains were partitioned. Even if the strains were divided into three different groups, with low, medium, or high rates of 16α-OHT production, none of the haplotype blocks identified had a strain grouping that was consistent with the rate of 16α-OHT generation. Therefore, the 23 haplotype blocks in which LP/J and 129/SvJ share a unique haplotype contain the most plausible candidates that explain the interstrain differences in the rate of 16α-OHT production. In fact, of the total genetic variance calculated using the mean rate of formation of this metabolite for each strain, 75% of the interstrain differences could be explained by the genetic difference within these 23 blocks.

Because of the strong genetic correlation with the pharmacokinetic data, genes within any of these 23 haplotype blocks could potentially be responsible for differential testosterone metabolism among the inbred strains. As shown in our previous pharmacogenetic (2) and other analyses (3, 5, 21), it is expected that many genomic regions will have patterns of genetic variation that will correlate with, but not be causative of, a measured phenotypic difference among the inbred strains. For this reason, we apply additional criteria to determine whether genetic differences within the correlated regions could cause the phenotype differences. These criteria enable the number of candidate genes among the genetically correlated regions to be quickly reduced. For an in vitro drug biotransformation reaction catalyzed by a liver extract, we first determine whether a candidate gene is expressed in liver and eliminate any regions that do not encode genes expressed in the liver. A gene expression database generated by microarray analysis of liver tissue obtained from 10 inbred strains was used for this analysis. If a gene was expressed in any of the three liver samples obtained from any one of the 10 strains analyzed, it was considered a candidate gene. These 23 haplotype blocks encoded a total of 24 genes, and only 17 were expressed in the liver according to this criterion (SI Table 3). Because a phase I drug biotransformation reaction catalyzed by a liver extract was analyzed, we evaluated whether any of these candidate genes encoded enzymes that could generate hydroxylated metabolites (22). The haplotype blocks encoding transcription factors or chromatin components (Hnf4a, Ttf1, and Smarcd1), ion channels (Kcnb1, Kcnb8, and Ktcd6), growth factors (Hgf and Eps8), matrix components (Sdc2), histocompatibility antigens (H2-T18), or receptors (Grin2a) were rapidly eliminated. Haplotype blocks encoding the following enzymes, which could not catalyze a hydroxylation reaction of this type, were also eliminated: proteases (Bace1, Adam10, and Prss), phosphatases (Ptpn21), and dehydrogenases (Gpd1). Therefore, the computationally identified 0.51-Mb region (21.06–21.57 Mb) on chromosome 7 that encodes three Cyp2b P450 enzymes (Cyp2b10, Cyp2b13, and Cyp2b9) was of particular interest (Fig. 2). There were two distinct haplotypes within this block, the LP/J and 129/SvJ strains, which shared a haplotype that differed from the other 13 strains (Fig. 2). Of note, >60 different Cyp450 enzymes, 5 flavin monooxygenases, 8 aldehyde dehydrogenases, and 41 nuclear hormone receptors (including those known to affect the expression of drug metabolizing enzymes) were genetically analyzed within the Roche SNP database. However, only these Cyp2b enzymes had a pattern of genetic variation that correlated with rate of 16α-OHT generation across the inbred strains. Because there was no overlap between the two genetically and phenotypically distinct strain groupings, this pattern of genetic variation had a very high degree of correlation with the pattern of metabolite generation (P value < 0.0001). The level of expression of Cyp2b9 and Cyp2b13 mRNA in liver was >10-fold higher than that of Cyp2b10. All three Cyp2b mRNAs were expressed at significantly lower levels in the two strains (LP/J and 129/SvJ) with a low rate of 16α-OHT formation than in the other 13 strains. Indeed, Cyp2b13 and Cyp2b9 mRNA was virtually absent in the 129/SvJ and LP/J strains (Fig. 3).

Fig. 2.

Fig. 2.

Open in a new tab

The location of the three Cyp2b genes (Cyp2b10, Cyp2b13, and Cyp2b9) on chromosome 7 and their haplotypes across 15 inbred mouse strains are shown. The chromosomal positions of these Cyp2b genes are indicated in base pairs downstream of the centromere at Left, and Right shows the structure of the haplotype block containing these three genes. The genomic position was determined using mouse genome National Center for Biotechnology Information (NCBI) build 34. Within a block, each column represents the indicated inbred mouse strain, and each box represents the corresponding allele for the indicated mouse strain. A blue box indicates that the strain has the common allele, whereas a yellow box indicates a minor allele, and an empty box indicates that the allele is unknown.

Fig. 3.

Fig. 3.

Open in a new tab

Cyp2b10, Cyp2b13, and Cyp2b9 mRNA expression in mouse liver. The amount of each Cyp2b mRNA expressed in liver prepared from females of 15 inbred mouse strains was measured by Taqman analysis. Each data point represents the average ± SD of three measurements performed on three independently obtained liver samples and was normalized relative to the level of Actin mRNA expression.

To identify the responsible enzyme, the in vitro rate of 16α-OHT formation catalyzed by expressed recombinant Cyp2b10, Cyp2b13, and Cyp2b9 enzymes prepared from a strain (DBA/2) with a high rate of 16α-OHT generation was measured. Cyp2b9 and Cyp2b10 catalyzed the conversion of testosterone to 16α-OHT in a testosterone concentration-dependent manner, whereas Cyp2b13 did not. Cyp2b9 mRNA was not expressed in the strains (129/SvJ and LP/J) with a low rate of 16α-OHT generation, but there was a low level of Cyp2b10 mRNA expression in these strains. However, expressed recombinant Cyp2b10 prepared from a strain (129) with a low rate of 16α-OHT formation could not mediate this biotransformation (Fig. 4). The activity of all of four recombinant enzymes, the DBA/2 and 129 allelic forms of Cyp2b9 and Cyp2b10, was also confirmed by assay of their P450 content and by their ability to biotransform testosterone to metabolites other than 16α-OHT (data not shown). DBA/2-derived recombinant Cyp2b9 has a significantly higher Vmax than Cyp2b10; more importantly, the measured Km for 16α-OHT formation by DBA/2-derived recombinant Cyp2b9 (0.6 μM) is >2 orders of magnitude lower than that of DBA/2-derived recombinant Cyp2b10 (101 μM) (Table 1). This suggests that the catalytic efficiency (Vmax/Km) of Cyp2b9 in mice with rapid 16α-OHT formation is ≈100-fold greater than that of Cyp2b10. Furthermore, Cyp2b9 mRNA is 10-fold more abundant than Cyp2b10 mRNA in liver (Fig. 3). Therefore, analysis of both catalytic activity and mRNA expression indicates that Cyp2b9 is the enzyme principally responsible for converting testosterone to 16α-OHT in mice (Table 1).

Fig. 4.

Fig. 4.

Open in a new tab

Cyp2b-mediated testosterone biotransformation. Cyp2b10, Cyp2b13, and Cyp2b9 cDNAs were generated from a strain with a high rate (DBA/2J) of 16α-OHT generation, and an additional allelic form of Cyp2b10 was amplified from a strain (129/SvJ) with a low rate of 16α-OHT generation. The rate of testosterone biotransformation into 16α-OHT by each expressed recombinant enzyme was measured after incubation with the indicated concentration of testosterone. The data are expressed as the rate (picomolar per minute per nanomolar CYP) of formation of 16α-OHT, and each data point represents the average ± SD of at least three individual measurements.

Table 1.

The kinetic parameters for 16-αhydroxytestosterone (16 α-OHT) production and SN-38 glucuronidation in vitro

Strain Enzyme Km Vmax Vmax/Km
DBA/2 Cyp2b9 0.6 ± 0.2 62.6 ± 9.9 106.1
DBA/2 Cyp2b10 101.4 ± 44.1 10.7 ± 3.3 0.1
DBA/2 Microsome 8.2 ± 0.7 1185.7 ± 74.5 144.6
C57/B6 Ugt1a7c 31.8 ± 12.1 28.8 ± 2.0 0.98
DBA/2 Ugt1a7c 19.8 ± 3.8 9.6 ± 1.3 0.48
C57/B6 S9 31.1 ± 8.3 0.26 ± 0.03 0.009
DBA/2 S9 25.1 ± 3.0 0.15 ± 0.01 0.006
Open in a new tab

The rate of 16α-OHT production was measured by using DBA/2J microsomes and expressed recombinant Cyp2b9 and Cyp2b10 (DBA/2J allelic forms) enzymes. The rate of SN-38 glucuronidation was measured by using C57/B6 and DBA/2J S9 and two different allelic forms (C57/B6 and DBA/2) of expressed recombinant Ugt1a7c. The Km and Vmax were calculated from measurements of the in vitro rate of 16a-OHT production or SN-38 glucuronidation catalyzed by hepatic microsomes or S9 extracts, or by the indicated expressed recombinant enzyme. The values are expressed as the rate (pmol/min per nmol CYP) of 16 α-OHT formation or as the rate of SN-38 glucuronidation (quantitated relative to the amount of parent drug). Each data point represents the average ± SD of three individual measurements.

We identified 155 SNPs within the Cyp2b9 gene locus and 150 SNPs in the 30-kb region surrounding Cyp2b10 among the 15 inbred strains analyzed. The Cyp2b9 SNPs did not alter the predicted amino acid sequence. Two SNPs within Cyp2b10 altered its amino acid sequence (Ile362Leu and Asp411Glu). All of the minor alleles in Cyp2b9 and Cyp2b10 are uniquely found in the two (LP/J and 129/SvJ) strains with a low rate of 16α-OHT formation. Thus, polymorphisms affecting Cyp2b9 mRNA expression in liver are principally responsible for the strain-specific differences in the 16α-OHT formation.

Testosterone 16α-hydroxylase activity in the liver has previously been shown to be regulated by an autosomal dominant sex-limited locus in inbred mice (11, 12). Analysis of the 2-kb promoter region 5′ of Cyp2b9 identified six potential estrogen receptor-binding sites (Transfac database, cutoff 0.9, P = 2e-08), which could explain the previously observed sex-dependent expression of the murine testosterone 16α-hydroxylase (12). Interestingly, Cyp2b9 is located within a chromosomal region that was identified >18 years ago by quantitative trait locus (QTL) analysis of the amount of hepatic testosterone 16α-hydroxylase activity present in female intercross progeny and in recombinant inbred strains (11, 12). However, the QTL mapping study required a very significant amount of time to generate and characterize drug metabolism in intercross progeny and identified a very large chromosomal region that may regulate the activity of this enzyme. In contrast, in vitro metabolite analysis coupled with haplotype-based computational mapping enabled individual candidate genes to be identified in several days (6).

Irinotecan Glucuronidation.

The enzymes mediating phase II drug biotransformation reactions have distinct properties and require different cofactors than the NADPH-dependent Cyp450 enzymes (23). Therefore, the protocol for preparation of liver extracts was modified to produce S9 extracts, and additional cofactors were added to enable phase II enzymes to be analyzed by this murine hepatic in vitro biotransformation system. S9 extracts from 15 inbred mouse strains were prepared, and their ability to catalyze the glucuronidation of the active metabolite of irinotecan (SN-38) was tested. The in vitro rate of SN-38 glucuronide formation (SN-38G) was quantitatively analyzed by LC/MS/MS analysis in three independently performed experiments for each strain. There was a highly reproducible 4-fold difference in the rate of SN-38G formation across the inbred strains (Fig. 5A). ANOVA indicated that the rate of SN-38G formation was significantly different among the inbred strains analyzed (P < 0.0001); 96.5% of the total variance in the three independent measurements obtained for each of the 15 strains was due to interstrain (genetic) differences, whereas within-strain variation accounted for only 3.5% of the total variance.

Fig. 5.

Fig. 5.

Open in a new tab

In vitro analysis of SN-38 glucuronidation. (A) The amount of SN-38 glucuronide formed after incubation (20 μM) of SN-38 with liver S9 fractions prepared from the indicated 15 inbred mouse strains was quantitated by LC/MS/MS analysis and plotted as a function of time. Each data point represents the average ± SD of results obtained from analysis of three individual incubation reactions. (B) The rate of glucuronidation was calculated by linear regression analysis (Upper) and then log-transformed for haplotype-based computational genetic analysis. Nine haplotype blocks that were most strongly correlated with the measured rate of glucuronidation are shown (Lower). For each predicted block, the chromosomal location, number of SNPs within a block, its gene symbol, and the level of mRNA expression in liver are shown. The haplotype for each strain is represented by a colored block, and the blocks are presented in the same order as the phenotypic data. The calculated P value measures the probability that the strain groupings within a block would have the same degree of association with the phenotypic data by random chance. The measured level of expression of the indicated gene using microarrays in liver tissue obtained from the strain with the highest measured value is shown. Each number is the average of three independent measurements for the strain with the highest level of expression, and liver tissue from 10 inbred strains was analyzed. “No Exp” indicates that the gene was not detectably expressed in any of the 10 strains examined.

Haplotype-based computational genetic analysis was used to identify haplotype blocks that correlated with the interstrain differences in the rate of SN-38 glucuronide formation. The nine haplotype blocks that best correlated are shown in Fig. 5B. The two haplotype blocks that were most strongly correlated (P < 0.0001) were within a genomic region (Ugt1a) on chromosome 1 that encoded several UGT enzymes that could mediate this glucuronidation reaction (Fig. 5B). Of note, almost all UDP glucuronyl transferases (n = 23), GST (n = 25), and sulfotransferase (n = 44) enzymes that can mediate a phase II biotransformation reaction were genetically analyzed within the Roche SNP database. However, Ugt1a was the only phase II enzyme within a computationally identified haplotype block. There are 274 SNPs within the Ugt1a locus among the 20 strains in our database, and these divided the 15 inbred mouse strains analyzed into two haplotypic groups. The eight strains with a low rate of SN-38 glucuronidation shared the same alleles within a haplotype block, whereas the seven strains with a high rate of glucuronidation had a unique haplotype (Fig. 6A). In addition, a few alleles that were uniquely present in the SMJ and NZB strains created a third haplotype within one of the two haplotype blocks located within the 168-kB Ugt1a gene.

Fig. 6.

Fig. 6.

Open in a new tab

Haplotype map and expressed isoforms of the murine Ugt1a1 gene. (A) Of 274 SNPs within the Ugt1a1 locus, the nine exonic SNPs that alter an amino acid in the predicted protein are shown. The 15 inbred strains have two haplotypes within the 168.2-kb Ugt1a1 locus on chromosome 1. However, polymorphisms in the SMJ strain created a third haplotype in part of this locus. Each column represents an inbred mouse strain, and each box indicates the corresponding allele for that strain, and the first column indicates the amino acid substitution caused by the polymorphism. The allele indicated by a yellow box corresponds to the amino acid indicated in red, whereas a blue box indicates the amino acid shown in blue. A gray box indicates an unknown allele is for that strain. (B) The five different mRNA isoforms encoded by nine exons within mouse Ugt1a locus. Each isoform is generated by alternative splicing of one of the first five exons, and all expressed isoforms share four common exons (exons 6–9). All nine SNPs that alter the amino acid sequence of the protein are located within the first two exons. Ugt1a9 mRNA contains exon 1, whereas Ugt1a7c mRNA contains exon 2.

Five different mRNAs (Ugt1a9, Ugt1a7c, Ugt1a6, Ugt1a2, and Ugt1a1) are produced through alternative splicing of the first five of the nine exons located within the Ugt1a gene (Fig. 6B). Nine SNPs altered the amino acid sequence of the protein, and all are located within the domain of the enzyme that catalyzes the transfer of uridine diphosphoglucuronic acid to a drug. Two amino acid alterations are within the first exon, which is uniquely present in the Ugt1a9 transcript; and the other 7-aa changes occur in the second exon that is uniquely found in the Ugt1a7c mRNA (Fig. 6B). No SNP caused an amino acid change in any of the other seven exons. We analyzed the level of expression of the five Ugt1a encoded mRNAs in liver obtained from the 15 inbred strains analyzed in the in vitro pharmacokinetic studies. Although there were differences in the level of expression of these mRNAs among the inbred strains, their pattern of expression did not correlate with the haplotype of the strain or with the measured rate of SN-38 glucuronidation (SI Fig. 9). Because differences in mRNA expression were not responsible, the SNPs causing amino acid substitutions within the Ugt1a locus could contribute to the observed strain-specific differences in the rate of SN-38 glucuronidation.

The nonsynonymous SNPs within the first and second exons of the Ugt1a gene would alter the predicted amino acid sequence of only two mRNAs, Ugt1a9 and Ugt1a7c. Therefore, we examined the ability of different allelic forms of expressed recombinant Ugt1a9 and Ugt1a7c enzymes to glucuronidate irinotecan. Neither allelic form of expressed recombinant Ugt1a9 catalyzed the glucuronidation of SN-38. As an additional specificity control, expressed recombinant Ugt3a2 also could not catalyze SN-38 glucuronidation. In contrast, both allelic forms of Ugt1a7c were able to glucuronidate this metabolite (Fig. 7). Consistent with the pharmacokinetic differences observed among the inbred strains, expressed recombinant Ugt1a7c from strains with a higher rate of glucuronidation had a significantly higher Vmax and catalytic efficiency (Vmax/Km) than the enzyme produced from strains with a lower rate of glucuronidation (Table 1). Of note, the measured Km for SN-38 glucuronidation for these expressed recombinant enzymes is similar to that measured in the S9 preparations. These results indicate that polymorphisms altering the amino acid sequence of Ugt1a7c contribute to the observed differences in the rate of irinotecan glucuronidation among inbred mouse strains.

Fig. 7.

Fig. 7.

Open in a new tab

The rate of SN-38 glucuronidation catalyzed by two different allelic forms (C57B6 and DBA/2) of expressed recombinant Ugt1a7c and Ugt1a9 enzymes. The in vitro rate of SN-38 glucuronidation was measured after 30 min of incubation of the indicated concentrations of SN-38 with these expressed recombinant proteins. As an additional specificity control, the ability of expressed recombinant Ugt3a2 to catalyze SN-38G formation was also tested. The amount of SN-38 glucuronide (SN-38G) is expressed as ratio of SN-38G formed relative to an internal standard. Each data point represents the average ± SD of three measurements, and there was no detectable activity for the expressed recombinant Ugt1a9 and Ugt3a2 enzymes.

The genetic differences responsible for interstrain differences in irinotecan metabolism were rapidly identified using this murine in vitro biotransformation system, and they mirror those responsible for a clinically important interindividual difference in patient response to this drug (13, 17, 18). Furthermore, the pattern of genetic variation within the murine Ugt1a gene is also similar to the polymorphisms associated with human hereditary hyperbilirubinemias (24). Polymorphisms in the TATAA box of the promoter (16), and a missense mutation (Gly71Arg) within the second exon (25) of the human UGT1A1 gene are associated with human hereditary hyperbilirubinemias. The large number of polymorphisms within the Ugt1a locus in mouse is also seen in humans. Analysis of a larger number of affected individuals indicates that several different types (and possibly combinations) of polymorphisms within the UGT1A1 locus may contribute to human hereditary hyperbilirubinemias (26, 27).

Although the findings confirm previously identified metabolic routes for the two drugs analyzed, they provide a proof of principle that some of the genetic factors responsible for interindividual differences in drug metabolism can be characterized by haplotype-based computational genetic analysis of data generated using an in vitro drug biotransformation system. In contrast to existing analysis methods that require large efforts and long time frames, the in silico and in vitro approach very rapidly identified the genetic basis for interindividual variability in drug metabolism. Two distinct types of drug biotransformation reactions (Cyp450 and UGT-mediated reactions) were successfully analyzed by this method, and these enzyme families contribute to the metabolism of a large number of drugs. Although these represent the only two drugs studied using this in vitro/in silico approach, that two drugs with such different metabolic profiles (phase I vs. phase II) were successfully analyzed suggests this approach may be broadly used to analyze a large number of current and future medications. Combining a powerful computational genetic analysis tool with a high-throughput in vitro drug biotransformation system produced a model system with the potential to transform how we identify and characterize interindividual variability in drug metabolism. Because drug-induced toxicities are often caused by specific metabolites, this system can also be used to analyze interindividual differences in drug response.

This particular in vitro method has limitations associated with using in vitro hepatic preparations. It cannot assess the effect of membrane transport on drug clearance. The metabolites generated in vitro may differ from those produced in vivo. Tissues other than liver may be involved in drug biotransformation. Despite these limitations, these examples demonstrate that utilization of this in silico and in vitro experimental genetic system has the potential to markedly accelerate pharmacogenetic analysis. More broadly, this approach can be adapted to probe every facet of murine physiology that can be studied using an in vitro system, simply by preparing the appropriate tissue extracts from multiple strains that are genetically characterized within our database.

Materials and Methods

Methods in SI Text.

The chemicals used and the methods for microsome and S9 extract preparation, in vitro analysis of drug metabolite formation, LC/MS/MS analysis, cDNA cloning and baculovirus expression, as well as microarray and RT-PCR gene expression analysis are described in SI Text.

Computational Genetic Mapping.

Haplotype-based computational genetic analysis of the pharmacokinetic data was performed as described (5–7). In brief, allelic data from multiple inbred strains were analyzed, and a haplotype block map of the mouse genome was constructed (28, 29). SNPs were organized into haplotype blocks. Only a limited number of haplotypes, typically two, three, or four, are present within a haplotype block. This analysis identifies haplotype blocks in which the haplotypic strain grouping within a block correlates with the distribution of phenotypic data among the inbred strains analyzed. To do this, a P value that assesses the likelihood that genetic variation within each block could underlie the observed distribution of phenotypes among the inbred strains is calculated as described by using ANOVA (5–7). The haplotype blocks are then ranked based on the calculated P value. The genomic regions within haplotype blocks that strongly correlated with the phenotypic data are then analyzed. The haplotype map had 5,255 haplotype blocks generated from 237,023 SNPs characterized across 21 inbred strains covering 2,882 genes. Polymorphisms within 60 Cyp450 enzymes, 8 aldehyde dehydrogenases, 5 flavin monooxygenases, 88 transporters (Abc or Slc gene families), almost all UDP glucuronyl transferase genes (23), GST (20), and sulfotransferases (44) were included in the Roche SNP database. Because several nuclear hormone receptors (Pxr, RxR, and Ahr) can affect the expression of drug metabolizing enzymes, this database also covers 41 nuclear hormone receptors. Because ANOVA-based computational analysis requires homogeneity of variance, the rate of metabolite formation for each strain was normalized by logarithmic transformation before the computational analysis was performed. For the testosterone analysis, the candidate haplotype blocks that were empirically selected had a P = 3.2 × 10−5. This was the best P value achieved by blocks in which the LP/J and 129/SvJ strains shared a haplotype that was distinct from other strains.

Supplementary Material

Supporting Information
pnas_0700724104_index.html (12.8KB, html)

Acknowledgments

We thank Bill Fitch and Hua-fen Liu for valuable discussions and Ezra Tai and Doug Clark for help with this manuscript. Y.G., P.L. and E.F. were supported by National Institute of General Medical Sciences Grant 1 R01 GM068885-01A1 (to G.P.).

Abbreviations

LC/MS/MS

liquid chromatography tandem MS

OHT

hydroxytestosterone

UGT

UDP-glucuronosyl transferase

SN-38

7-ethyl-10-hydroxy-camptothecin.

Footnotes

Conflict of interest statement: Y.G., P.L., E.F., X.Z., J.W., G.L., Z.Z., S.H., J.A., J.U., M.M., and G.P. are employees of Roche Palo Alto.

This article is a PNAS Direct Submission.

This article contains supporting information online at www.pnas.org/cgi/content/full/0700724104/DC1.

References

  • 1.Guo Y, Shafer S, Weller P, Usuka J, Peltz G. Pharmacogenomics. 2005;6:857–864. doi: 10.2217/14622416.6.8.857. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Guo YY, Weller PF, Farrell E, Cheung P, Fitch B, Clark D, Wu S-Y, Wang J, Liao G, Zhang Z, et al. Nat Biotechnol. 2006;24:531–536. doi: 10.1038/nbt1195. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Liang D, Liao G, Wang J, Usuka J, Guo YY, Peltz G, Clark JD. Anesthesiology. 2006;104:1054–1062. doi: 10.1097/00000542-200605000-00023. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Liang D-Y, Liao G, Lighthall G, Peltz G, Clark JD. Pharmacogenet Genomics. 2006;16:825–835. doi: 10.1097/01.fpc.0000236321.94271.f8. [DOI] [PubMed] [Google Scholar]
  • 5.Liao G, Wang J, Guo J, Allard J, Chang J, Nguyen A, Shafer S, Puech A, McPherson JD, Foernzler D, et al. Science. 2004;306:690–695. doi: 10.1126/science.1100636. [DOI] [PubMed] [Google Scholar]
  • 6.Wang J, Liao G, Usuka J, Peltz G. Trends Genet. 2005;21:526–532. doi: 10.1016/j.tig.2005.06.010. [DOI] [PubMed] [Google Scholar]
  • 7.Wang J, Peltz G. In: Computational Genetics and Genomics: New Tools for Understanding Disease. Peltz G, editor. Totowa, NJ: Humana; 2005. pp. 51–70. [Google Scholar]
  • 8.Cervino AC, Li G, Edwards S, Zhu J, Laurie C, Tokiwa G, Lum PY, Wang S, Castellini LW, Lusis AJ, et al. Genomics. 2005;86:505–517. doi: 10.1016/j.ygeno.2005.07.010. [DOI] [PubMed] [Google Scholar]
  • 9.Yu J, Pressoir G, Briggs WH, Vroh Bi I, Yamasaki M, Doebley JF, McMullen MD, Gaut BS, Nielsen DM, Holland JB, et al. Nat Genet. 2006;38:203–208. doi: 10.1038/ng1702. [DOI] [PubMed] [Google Scholar]
  • 10.Petkov P, Graber J, Churchill G, DiPetrillo K, King B, Paigen K. PLoS Genet. 2005;1:e33. doi: 10.1371/journal.pgen.0010033. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Noshiro M, Negishi M. Biochemistry. 1988;27:6444–6448. doi: 10.1021/bi00417a036. [DOI] [PubMed] [Google Scholar]
  • 12.Noshiro M, Lakso M, Kawajiri K, Negishi M. Biochemistry. 1988;27:6434–6443. doi: 10.1021/bi00417a035. [DOI] [PubMed] [Google Scholar]
  • 13.Gagne JF, Montminy VBP, Gaucher G, Guillemette C. Mol Pharmacol. 2002;62:608–617. doi: 10.1124/mol.62.3.608. [DOI] [PubMed] [Google Scholar]
  • 14.Mackenzie PI, Bock KW, Burchell B, Guillemette C, Ikushiro S, Iyanagi T, Miners JO, Owens IS, Nebert DW. Pharmacogenet Genomics. 2005;15:677–685. doi: 10.1097/01.fpc.0000173483.13689.56. [DOI] [PubMed] [Google Scholar]
  • 15.Nagar S, Blanchard RL. Drug Metab Rev. 2006;38:393–409. doi: 10.1080/03602530600739835. [DOI] [PubMed] [Google Scholar]
  • 16.Bosma PJ, Chowdhury JR, Bakker C, Gantla S, de Boer A, Oostra BA, Lindhout D, Tytgat GNJ, Jansen PLM, Elferink RPJ, et al. N Engl J Med. 1995;333:1171–1175. doi: 10.1056/NEJM199511023331802. [DOI] [PubMed] [Google Scholar]
  • 17.Ishikawa K, Kajita Y, Hasegawa Y, Noda Y, Yoshida J, Nabeshima T. J Neurooncol. 2005;74:283–286. doi: 10.1007/s11060-004-7119-4. [DOI] [PubMed] [Google Scholar]
  • 18.Toffoli G, Cecchin E, Corona G, Russo A, Buonadonna A, D'Andrea M, Pasetto LM, Pessa S, Errante D, De Pangher V, et al. J Clin Oncol. 2006;24:3061–3068. doi: 10.1200/JCO.2005.05.5400. [DOI] [PubMed] [Google Scholar]
  • 19.Isidori A, Giannetta E, Gianfrilli D, Greco E, Bonifacio V, Aversa A, Isidori A, Fabbri A, Lenzi A. Clin Endocrinol. 2005;63:381–394. doi: 10.1111/j.1365-2265.2005.02350.x. [DOI] [PubMed] [Google Scholar]
  • 20.Ford HC, Wheeler R, Engel LL. Eur J Biochem. 1975;57:9–14. doi: 10.1111/j.1432-1033.1975.tb02271.x. [DOI] [PubMed] [Google Scholar]
  • 21.Cashman JR, Park SB, Yang ZC, Washington CB, Gomez DY, Giacomini KM, Brett CM. Drug Metab Dis. 1993;21:587–597. [PubMed] [Google Scholar]
  • 22.Crespi CL, Miller VP. Pharmacol Ther. 1999;84:121–131. doi: 10.1016/s0163-7258(99)00028-5. [DOI] [PubMed] [Google Scholar]
  • 23.Brandon E, Raap C, Meijerman I, Beijnen J, Schellens J. Toxicol Appl Pharmacol. 2003;189:233–246. doi: 10.1016/s0041-008x(03)00128-5. [DOI] [PubMed] [Google Scholar]
  • 24.Chowdhury JR, Wolkoff AW, Chowdhury NR, Arias IM. In: The Metabolic and Molecular Bases of Inherited Disease. Scriver CR, Beaudet AL, Sly WS, Vale D, editors. New York: McGraw–Hill; 2001. pp. 3063–3101. [Google Scholar]
  • 25.Moghrabi N, Clarke DJ, Boxer M, Burchell B. Genomics. 1993;18:171–173. doi: 10.1006/geno.1993.1451. [DOI] [PubMed] [Google Scholar]
  • 26.Petit F, Gajdos V, Capel L, Parisot F, Myara A, Francoual J, Labrune P. Clin Genet. 2006;69:525–527. doi: 10.1111/j.1399-0004.2006.00616.x. [DOI] [PubMed] [Google Scholar]
  • 27.Yamamoto K, Soeda Y, Kamisako T, Hosaka H, Fukano M, Sato H, Fujiyama Y, Adachi Y, Satoh Y, Bamba T. J Hum Genet. 1998;43:111–114. doi: 10.1007/s100380050050. [DOI] [PubMed] [Google Scholar]
  • 28.Wang J, Liao G, Chang J, Nguyen A, Guo J, Chou S, Hu S, Bach C, Allard J, Shafer S, et al. In: Computational Genetics and Genomics: New Tools for Understanding Disease. Peltz G, editor. Totowa, NJ: Humana; 2005. pp. 71–83. [Google Scholar]
  • 29.Zorzano A, Herrera E. Biochem Pharmacol. 1990;39:873–878. doi: 10.1016/0006-2952(90)90202-v. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supporting Information
pnas_0700724104_index.html (12.8KB, html)
pnas_0700724104_00724Fig8a.jpg (25.4KB, jpg)
pnas_0700724104_00724Fig8b.jpg (30.2KB, jpg)
pnas_0700724104_00724Fig9.jpg (51KB, jpg)

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences

RESOURCES