Estrogens, Enzyme Variants, and Breast Cancer: A Risk Model

Numerous epidemiologic studies have implicated estrogens in the development of breast cancer (1, 2). The two major estrogens, 17β-estradiol (E₂) and estrone (E₁), are ligands for the estrogen receptor and substrates for oxidizing phase I enzymes, cytochrome P450 (CYP) 1A1 and CYP1B1. In their dual role of ligand and substrate, estrogens may simultaneously stimulate cell proliferation and gene expression via the estrogen receptor and cause DNA damage via their oxidation products, the catechol estrogens (3, 4). The latter mechanism is based on the unique chemical structure of estrogens. Unlike all other steroid hormones, estrogens have an aromatic A-ring, which yields catechols on oxidation that may be further oxidized to highly reactive semiquinones and quinones (Fig. 1), which in turn can form both oxidative and estrogen DNA adducts. Thus, estrogen quinones seem to share a common feature of many chemical carcinogens (i.e., the ability to covalently modify DNA; refs. 5-8). Support for the carcinogenic activity of estrogens and their oxidative products, the catechol estrogens, comes from experiments in animal models. Treatment with E₂ and the catechol 4-hydroxyestrogen and 2-hydroxyestrogen caused kidney cancer in male Syrian hamsters and endometrial cancer in female CD1 mice (9-11). However, there is no animal model for estrogen-induced breast cancer, and even in the hamster and mouse models, the precise mechanism of DNA damage is uncertain. Thus, there is a need to understand estrogen metabolism in the human breast to elucidate the role of endogenous and exogenous estrogens in mammary carcinogenesis. To advance this understanding requires not only characterization of the various estrogen metabolites but also equally important a precise definition of the responsible enzymes.

Several investigators have proposed a qualitative model of mammary estrogen metabolism regulated by oxidizing phase I and conjugating phase II enzymes (12, 13). The oxidative estrogen metabolism pathway starts with E₂ and E₁, which are oxidized to the 2-OH and 4-OH catechol estrogens by the phase I enzymes CYP1A1 and CYP1B1 (14, 15). These same enzymes are postulated to further oxidize the catechol estrogens to unstable semiquinones and quinones. Estrogen quinones then form Michael addition products with deoxynucleosides (5, 6, 16). The catechol estrogens and their estrogen quinones/semiquinones also undergo redox cycling, which results in the production of reactive oxygen species capable of causing oxidative DNA damage (17-19). Thus, P450-mediated estrogen metabolism is expected to lead to the formation of both oxidative and estrogen DNA adducts, all of which have been shown to possess mutagenic potential (20-22). It is postulated that the genotoxicity of the oxidative estrogen metabolism pathway is mitigated by alternate reactions of the metabolites with phase II enzymes. Specifically, catechol-O-methyltransferase (COMT) catalyzes the methylation of catechol estrogens to methoxy estrogens, which lowers the catechol estrogens available for conversion to estrogen quinones (23, 24). In turn, the estrogen quinones undergo conjugation with reduced glutathione (GSH) via the catalytic action of glutathione S-transferase (GST) P1 (25, 26). The formation of GSH-estrogen conjugates would reduce the level of estrogen quinones and thereby lower the potential for DNA damage.

The current model of mammary estrogen metabolism has several limitations. First, only single enzymes (e.g., CYP1B1 and COMT) have been analyzed to date with simple substrate-product kinetics, which clearly generates an incomplete picture of the metabolic pathway. Second, although the model incorporates the functional roles of the phase I and II enzymes, it remains uncertain how the enzymes interact quantitatively. Third, each of the phase I and II enzymes contains genetic polymorphisms (15, 23, 27, 28). Studies from several laboratories, including our own, have examined the functional implications of the polymorphisms on estrogen metabolism, again focusing on single enzymes (15, 23, 24, 29, 30). Thus, the multitude of potential kinetic reactions resulting from the complex genetic variations of the phase I and II enzymes is completely outside the scope of the current model of estrogen metabolism. We developed recently an experimental in vitro model of mammary estrogen metabolism, in which we used purified, recombinant phase I enzymes CYP1A1 and CYP1B1 with the phase II enzymes COMT and GSTP1 to determine how E₂ is metabolized (31). We used both gas and liquid chromatography with mass spectrometry (GC/MS and LC/MS) to measure the parent hormone E₂ as well as eight metabolites (i.e., the catechol estrogens, methoxyestrogens, and estrogen-GSH conjugates; Fig. 1). In this article, we used these experimental data to develop a multicompartmental kinetic model of the metabolic pathway. Furthermore, we used previously determined rate constants of variant CYP1A1, CYP1B1, and COMT to present an in silico kinetic-genomic model of mammary estrogen metabolism. Finally, we applied the model to a breast cancer case-control population and determined that the combination of enzyme haplotypes with E₂-3,4-Q production did indeed identify a subset of women with increased breast cancer risk.

We developed a mathematical model for the estrogen metabolism pathway shown in Fig. 1. We assume that each reaction in the pathway (A → B, a generic step in the pathway) is an enzyme-catalyzed reaction of the form: , where E denotes the enzyme, C is the enzyme-substrate complex, and k_i, i = 1, 2, and 3, are the rate constants of the reaction. For these types of reaction, we approximate the kinetics using the quasi steady-state assumption:

There are parts of the pathway for which kinetic data are not available. In particular, rate constants cannot be determined experimentally for the reaction sequences 2-hydroxyestradiol (2-OHE₂) → E₂-2,3-SQ → 2-OHE₂-quinone (E₂-2,3-Q) and 4-OHE₂ → E₂-3,4-SQ → E₂-3,4-Q because of the transient nature of the semiquinones (ms half-life; ref. 32). Therefore, we simplified the pathway and collapsed the sequential reactions to single reactions, 2-OHE₂ → E₂-2,3-Q and 4-OHE₂ → E₂-3,4-Q, respectively. We also assumed that each of these quinone production reactions (OHE₂^k → EQ₂^ij) satisfies dynamics of the form

As a measure of the quinone concentrations over the course of time, we introduce the area under the curve (AUC) metric:

Wild-type (WT) CYP1A1 cDNA was prepared for expression and purification of recombinant CYP1A1 as described previously (15, 33). Site-directed mutagenesis was done to generate the cDNA variants, which were verified by nucleotide sequence analysis and then similarly expressed and purified: 462Ile → Val (m2), 461Thr → Asn (m4), and 461Asn-462Val (m2/m4) (27). SDS-PAGE showed >95% protein purity and the reduced-CO difference spectrum revealed the λ_max at 450 nm, which allowed quantitation for subsequent enzyme experiments. We used GC/MS (15, 33) to determine the reaction kinetics of E₂ oxidation for WT, m2, m4, and m2/m4 CYP1A1.

The hospital-based case-control study group of 221 Caucasian women with primary invasive breast cancer and their age-matched control subjects has been described previously (34-36). Genomic DNA was extracted from tumor tissue or WBCs. The DNA samples of four control subjects had been depleted, leaving 221 cases and 217 controls for the study group.

The genotypes of CYP1A1, CYP1B1, and COMT were determined by PCR and restriction endonuclease digestion as described previously (23, 35, 36). Each PCR contained internal controls for the respective gene and random retesting of 50 samples yielded 100% reproducibility. Direct sequencing of five different samples provided further independent genotype validation.

The Wilcoxon rank sum test was used to determine the median difference in age between cases and controls. The χ² test was used to compare the distribution of CYP1A1, CYP1B1, and COMT alleles in cases and controls. The χ² goodness-of-fit test was used for testing Hardy-Weinberg equilibrium (HWE). Haplotype frequencies were estimated via the expectation-maximization algorithm in Powermarker version 3.23 (37, 38). Haplotype-trait association with breast cancer was tested using a regression approach also in Powermarker version 3.23 (37, 39).

There are 4 CYP1A1, 16 CYP1B1, and 2 COMT (23, 35, 36) haplotypes with 4 × 16 × 2 = 128 possible genetic combinations. We calculated an E₂-3,4-Q AUC for each woman based on her CYP1A1, CYP1B1, and COMT haplotypes, which together we term a composite haplotype. In calculating each AUC, we considered that the CYP1B1 gene has four polymorphic sites with 16 possible haplotypes. Because any individual can possess only two haplotypes, the certainty of assigning the correct CYP1B1 haplotypes becomes a matter of probability if the individual is heterozygous for more than one polymorphic site. The probability of a particular composite haplotype occurring in our population of 438 women was computed in the following manner. For each individual, a subset of the 128 possible combinations was computed resulting in 438 subsets of haplotype combinations (CYP1A1, CYP1B1, and COMT). Then, a count was done for each of the 128 possible composite haplotypes in the 438 subsets from which a frequency chart was constructed. From this chart, we defined the probabilities of composite haplotypes occurring in the population. Suppose an individual has n possible composite haplotypes of CYP1A1, CYP1B, and COMT and let AUC₁, AUC₂,…AUC_n denote the AUC values for each haplotype. If the probabilities that these composite haplotypes occur in the population are P₁, P₂,…P_n, where 0 < P_i < 1, i = 1, 2…n, then the composite AUC_comp is defined as

In a previous study, we determined the metabolism of E₂, 2-OHE₂, 4-OHE₂, 2-methoxyestradiol (2-MeOE₂), 2-hydroxy-3-methoxyestradiol (2-OH-3-MeOE₂), 4-methoxyestradiol (4-MeOE₂), 2-OHE₂-1-SG, 2-OHE₂-4-SG, and 4-OHE₂-2-SG as a function of time in the presence of CYP1A1 (85 pmol), CYP1B1 (165 pmol), COMT (125 pmol), and GSTP1 (500 pmol). Each experimental reaction contained 10 μmol/L E₂, 100 μmol/L S-adenosyl methionine, 100 μmol/L glutathione and proceeded for 0, 2, 5, 10, 20, and 30 minutes at 37°C followed by GC/MS and LC/MS analysis (31). Figure 2A shows superimposed the experimental data (dots) and the model simulations (curves) for all nine analytes over the 30-minute reaction time. In the simulations, it was assumed that initially all analyte concentrations are zero, except E₂ (0) = E₂^*. Enzyme concentrations used in the simulations are consistent with those used in the preceding experimental studies (15, 23, 26, 31, 33). Given the complexity of the pathway, there is excellent agreement between the simulated and experimental results. Of the nine analytes, only two, 2-MeOE₂ and 2-OH-3-MeOE₂, showed a noticeable difference between simulated and measured results. As shown in Fig. 1, the likely reason for this discrepancy lies in the more complex kinetics of 2-MeOE₂ and 2-OH-3-MeOE₂. These methoxyestrogens are the only analytes that are subject to the simultaneous action of three enzymes (i.e., the COMT-mediated production, which is counteracted by CYP1A1- and CYP1B1-mediated demethylation; ref. 33).

The estrogen quinones are too labile to be reliably quantified in a multienzyme system. However, as outlined in Materials and Methods, we could use the mathematical model to provide functional relations between E₂(t) and the estrogen quinone concentrations: EQ₂²³(t) and EQ₂³⁴(t). Figure 2B shows the simulated production and disappearance of the estrogen quinones during the 30-minute reaction with a lower level and faster, nearly complete disappearance of E₂-2,3-Q compared with the higher, more sustained level of E₂-3,4-Q.

Because the model was built on experimentally determined rate constants, we could analyze how variations in these kinetic variables, occurring as the result of enzyme polymorphisms, affect single steps or a combination of steps in the pathway. Table 1 summarizes the kinetic variables determined for variants of CYP1A1 (this study), CYP1B1 (40), and COMT (23). We applied the model to simulate reactions using the rate constants for 4 CYP1A1, 16 CYP1B1, and 2 COMT WT and variant enzymes. For each of the 4 × 16 × 2 = 128 possible genetic combinations, we simulated values for the resulting estrogen metabolites over the 30-minute reaction time and then used interpolatory polynomials for these functions to calculate the respective AUCs. These simulations permitted us to see that the combinations of enzyme variants produced a continuous spectrum of concentrations over time for each of the estrogen metabolites. Accordingly, the model allowed us to identify which variant combinations of CYP1A1, CYP1B1, and COMT produced the highest or lowest estrogen metabolite concentrations over time. Because the catechols and quinones have been shown to cause DNA damage, we focused our attention on these two groups of metabolites (Fig. 3). Of the 128 combinations of CYP1A1, CYP1B1, and COMT variants, we found, for example, that the haplotype combination CYP1A1_{461Asn-462Ile}CYP1B1_{48Arg-119Ser-432Val-453Asn}COMT_108Met produced the maximum AUC for both 4-OHE₂ and E₂-3,4-Q (Fig. 3B and D).

We applied the model to a hospital-based breast cancer case-control population that has been analyzed previously (34-36). The two principal differences to the preceding studies are (a) the evaluation of haplotypes rather than genotypes and (b) the integrated examination of CYP1A1, CYP1B1, and COMT instead of as independent entities. Table 2 summarizes the allele and haplotype data of the case-control population. Only 3 of the 4 possible CYP1A1 haplotypes and 12 of 16 possible CYP1B1 haplotypes were observed in the study group. Among the 12 CYP1B1 haplotypes, 8 were present in both cases and controls. Three of the uncommon haplotypes were seen only in cases and one rare haplotype was found only in controls. The overall P for the CYP1B1 haplotype distribution among cases and controls was 0.63. In the simulations, we focused our analysis on the E₂-3,4-Q AUC because E₂-3,4-Q has been identified as the principal estrogen metabolite causing DNA damage (8-10, 41). We calculated a composite E₂-3,4-Q AUC for each woman based on her CYP1A1, CYP1B1, and COMT haplotypes as outlined in Materials and Methods. This information was then used to rank every woman in the entire study population based on her individual E₂-3,4-Q AUC (Fig. 4).

A major weakness of genetic studies is the neglect of phenotypic factors. This is particularly true for polymorphic enzymes whose activity levels can vary considerably more in response to inducing agents than as a result of a single inherited amino acid substitution. For this reason, we considered the effect of changing the concentration of the phase I enzymes, which play the principal role in the metabolic pathway. As shown in Fig. 4, we varied the CYP1B1/CYP1A1 ratio from 2 to 5 in the model to reflect reported 4-OHE₂/2-OHE₂ ratios in breast tissue (42-44). In these simulations, we changed the concentrations of CYP1B1 while keeping CYP1A1 constant. The concentrations of COMT and GSTP1 remained unchanged. When we ranked the E₂-3,4-Q AUCs for the entire study population at different CYP1B1/CYP1A1 ratios, we observed an increase in median AUCs for cases and controls with increasing CYP ratio (Fig. 4A). There were no significant differences between case and control E₂-3,4-Q AUCs at any CYP1B1/CYP1A1 ratio. However, cases predominated in the top tier of the population as shown for the top 8 percentile (35 subjects) of the study group (Fig. 4B). At CYP1B1/CYP1A1 = 5, the model identified 23 cases and 12 controls (P = 0.06). The discriminating ability was even more pronounced in the top 2 percentile (10 subjects) of the study population (Fig. 4C). At CYP1B1/CYP1A1 = 5, there were nine cases and 1 control (P = 0.01). Table 3 summarizes the composite CYP1A1, CYP1B1, and COMT haplotypes together with the E₂-3,4-Q AUCs for the top 10 women.

The complexity of mammary estrogen metabolism was recognized several years ago and outlined in a qualitative model (12, 13). Although this model defined the role of specific components (e.g., the oxidizing phase I and conjugating phase II enzymes), the quantitative effect of these enzymes in the overall pathway could not be assessed. The experimental analysis of single enzymes with simple substrate-product kinetics offered an incomplete picture of the pathway limited to the enzyme examined. Here, we present a novel approach that incorporates experimental data obtained previously with individual enzymes into a mathematical model of the estrogen metabolism pathway. Instead of simply doing a parametric fitting exercise, we used actual experimental rate constants to develop the model, which consists of 11 differential equations that permit us to simulate the kinetics of E₂ and 8 metabolites in the multienzyme pathway. The model simulations were validated against experimental results obtained previously by incubating E₂ with the combined enzymes CYP1A1, CYP1B1, COMT, and GSTP1 (31) and showed excellent concordance of simulated and measured results. Of the nine analytes, only 2-MeOE₂ and 2-OH-3-MeOE₂ showed a noticeable deviation of the simulated from the measured results, most likely due to their more complex kinetics resulting from the simultaneous involvement of three enzymes, COMT, CYP1A1, and CYP1B1. It is noteworthy that the deviation of the two methoxyestrogens did not affect the simulation of more distal metabolites, such as the GSH-estrogen conjugates, which showed excellent agreement (Fig. 2A).

Catechol estrogens and estrogen quinones occupy pivotal positions in the oxidative estrogen metabolism pathway (Fig. 1). Using GC/MS, we could follow the production and disappearance of the catechol estrogens (Fig. 2A). Ideally, measurements of the estrogen quinones should be made, but they are highly reactive with short half-lives (seconds to minutes) due to the strained 1,2-diketone functionality inherent in o-quinones (45). Although estrogen quinones are too labile to be reliably quantified in a multienzyme system, the model allowed us to simulate their production and disappearance during the 30-minute reaction (Fig. 2B). The disappearance of the quinones is due to two factors, the conversion of the quinones into stable GSH-estrogen conjugates and the irreversible loss from the system, most likely due to binding of the reactive quinones to protein (46). The more rapid disappearance of E₂-2,3-Q compared with E₂-3,4-Q is consistent with the shorter half-life of 42 seconds for E₁-2,3-Q compared with 12 minutes for E₁-3,4-Q (47). Overall, the model captures the joint action of the phase I and II enzymes rather well, allowing the simulation of the pathway from the parent hormone E₂ through several enzymatic steps to the most distal metabolites, the GSH-estrogen conjugates. Because these conjugates are produced via the quinones, the excellent agreement between simulated and measured GSH-estrogen conjugate levels provides further assurance about the validity of modeling the estrogen quinones.

Although other phase I enzymes, such as CYP1A2 and CYP3A4, are involved in hepatic and extrahepatic estrogen oxidation, CYP1A1 and CYP1B1 display the highest levels of expression in breast tissue and therefore are considered the principal oxidizing enzymes in mammary estrogen metabolism (48, 49). COMT shows ubiquitous expression in all tissues, including breast (50). Although COMT is the sole methylating enzyme, there are potentially three GSH-conjugating enzymes active in the pathway. Based on protein levels in breast tissue, GSTP1 is the predominant member of the GST family with GSTM1 and GSTA1 present at much lower levels (51-53). GSTs are known to have selective as well as overlapping substrate specificities and it is presently not known whether GSTM1 and GSTA1 share with GSTP1 the ability to conjugate estrogen quinones (26). To determine the potential roles of GSTM1 and GSTA1 in estrogen metabolism, we plan to prepare each as purified, recombinant enzyme followed by kinetic studies to define their respective rate constants. Besides COMT, there are two other classes of phase II enzymes capable of conjugating catechol estrogens (i.e., the sulfotransferases and UDP-glucuronosyltransferases). It seems that the catechol estrogens are converted predominantly to methyl conjugates and to a lesser extent to sulfate and glucuronide conjugates (54). In future experiments, we will assess the role of sulfotransferases and UDP-glucuronosyltransferases. The present mathematical model only incorporates the key phase I enzymes CYP1A1 and CYP1B1 and the phase II enzymes COMT and GSTP1. However, the model can readily accommodate additional enzymes and allow inclusion of other GST members as well as sulfotransferases and UDP-glucuronosyltransferases in the same manner as we currently do for CYP1A1 and CYP1B1. In contrast to the complex kinetics of the methoxyestrogens, the sulfate and glucuronide conjugation reactions follow simple substrate-product kinetics like the GSTP1-mediated GSH conjugation. Therefore, we anticipate straightforward modeling with good agreement of simulated and experimental data.

Each of the phase I and II enzymes involved in estrogen metabolism possesses genetic variants that (a) are associated with altered enzyme function and (b) occur in a sizeable portion of the population (55, 56). Because our experimental analysis used only WT recombinant enzymes, the results provide a limited view of estrogen metabolism. To obtain a more realistic and inclusive view of estrogen metabolism in the female population, we used the mathematical model to simulate how variations in the kinetic variables resulting from polymorphisms of the enzymes affect the metabolite concentrations. We examined 4 CYP1A1, 16 CYP1B1, and 2 COMT alleles. GSTP1 also has two polymorphisms (i.e., 104Ile → Val and 113Ala → Val; refs. 28, 57), but it is unknown whether they affect GSH-estrogen conjugation. Thus, our simulations are based on the examination of 4 × 16 × 2 genetic combinations to show the utility of the model. Although each of the metabolites can be modeled, we concentrated our analysis on the catechols and quinones because of their documented carcinogenic activity (5-7, 16). As shown in Fig. 3, modeling of the 128 haplotype combinations produced a continuous spectrum of catechol and quinone concentrations over time, as expressed by a range of AUCs. The simulations identified the haplotype combinations producing the highest and lowest AUCs. For example, the maximum AUCs for 4-OHE₂ and E₂-3,4-Q were produced by the haplotype CYP1A1_{461Asn-462Ile}CYP1B1_{48Arg-119Ser-432Val-453Asn}COMT_108Met, which were 2.6- and 4.6-fold higher, respectively, than the minimum AUCs produced by haplotype CYP1A1_{461Thr-462Val}CYP1B1_{48Gly-119Ala-432Val-453Ser}COMT_108Val. Although these differences may not appear large, it is important to consider that they affect on lifetime exposure, which is consistent with the hormonal risk model presented by Pike et al. (58).

Our kinetic-genomic model is pertinent to the numerous epidemiologic studies that have examined the association of genetic variants of enzymes involved in estrogen metabolism with breast cancer risk (56, 59). These studies have been handicapped by investigating only one or two enzymes, but even those examining all enzymes have been fundamentally limited by not being able to assess the underlying metabolic interactions (60). Our model attempts to fill this gap and we applied it to a hospital-based case-control population that has been analyzed previously with respect to CYP1A1, CYP1B1, and COMT genotypes (34–36). Here, we went beyond genotypes and used the model to determine for each woman the effect of her composite CYP1A1, CYP1B1, and COMT haplotypes on estrogen metabolite production. Inherited variations in enzyme genotype persist throughout life and can therefore be regarded as constants for each individual. However, the very same genes are also subject to induction and levels of enzyme expression may vary considerably as a result of the high degree of inducibility by a variety of agents. For example, CYP enzymes are induced by hundreds of compounds (dietary and environmental chemicals and drugs) and human exposure to such xenobiotics is unavoidable (61). Intraindividual and interindividual variation in xenobiotic exposure has several consequences: (a) the P450 activity in an individual may change over time, (b) the P450 activity may differ between individuals of the same genotype, and (c) the phenotypic variability in P450 activity may be greater than the effect of genetic polymorphisms due to the strong inducing power of certain xenobiotics. Thus, although each individual has a unique composite E₂-3,4-Q AUC based on her subset of 128 genetic combinations, the AUC value can vary with the phenotype. In the model, we attempted to incorporate both the certainty of the enzyme genotype and the ambiguity of the phenotype, the latter indicated by the changing ratio of phase I enzymes CYP1B1/CYP1A1 (Fig. 4). Although we have to accept imprecise information about P450 activity in breast tissue, we can assume that the concentration of CYP1B1 is greater than that of CYP1A1 based on mRNA expression levels, higher levels of 4-OHE₂ than 2-OHE₂, and the observation that 2-OHE₂ is produced by both CYP isoforms, whereas 4-OHE₂ is formed only by CYP1B1 (14, 15, 48, 49). Because the 4-OHE₂/2-OHE₂ ratio can be ∼3 and reach as high as ∼5 (42-44), we varied the CYP1B1/CYP1A1 ratio in the model from 2 to 5. For CYP ratios >2, the model identified a top tier of E₂-3,4-Q AUCs with significantly increased numbers of breast cancer cases in the top percentiles (Fig. 4B and C; Table 3), suggesting that E₂-3,4-Q AUC may be an indicator of breast cancer risk. The ranking order of E₂-3,4-Q AUCs is primarily determined by the enzyme genotype (i.e., the composite CYP1A1-CYP1B1-COMT haplotype of a subject). However, the AUC ranking is also affected by the enzyme phenotype and a change in CYP1B1/CYP1A1 ratio may lead to a different ranking of a subject in the population (Fig. 4B and C). This is due to the fact that CYP1B1 and CYP1A1 catalyze different reactions in the metabolic pathway. Changing their ratio will have different results on the E₂-3,4-Q AUC for subjects with different composite haplotypes. Estrogens have long been recognized as prime risk factor for the development of breast cancer, but their assessment has not progressed beyond traditional exposure data, such as parity, age at menarche and menopause, etc. Here, we present a novel approach that is based on the molecular analysis of mammary estrogen metabolism. The E₂-3,4-Q AUC is a plausible metabolic risk factor for breast cancer because E₂-3,4-Q has been identified as principal estrogen metabolite causing DNA adduct formation in experimental animals and E₂-3,4-Q-derived DNA adducts have been detected in human breast cancer tissues (8-10, 13, 41, 44, 62). Whether the E₂-3,4-Q AUC is an independent risk factor, as suggested by the present analysis, will need to be confirmed by a larger separate study. The value of E₂-3,4-Q AUC as a new metabolic-genetic risk factor may yet be in its combination with traditional measures of endogenous and exogenous estrogen exposure. For example, one could estimate the overall exposure of a woman to E₂-3,4-Q AUC by taking into account (a) total years of menstruation or menopause age, (b) total pregnancy time, (c) years of menstruation before first full-term pregnancy, (d) body mass index, (e) dosage and duration of oral contraceptives, and (f) dosage and duration of hormone replacement therapy. Altogether, one could derive an individualized risk factor of estrogen exposure for each woman that combines her reproductive life history with her unique genetic and metabolic traits. Data on traditional variables related to estrogen exposure were unfortunately not obtained for all subjects of the present study population, such as the control subject who had the highest E₂-3,4-Q AUC in the entire population (Table 3).

In summary, using experimentally determined rate constants, we developed a mathematical model of mammary estrogen metabolism that allowed the kinetic simulation of E₂ and eight metabolites. The simulations showed excellent agreement with experimental results and provided a quantitative assessment of the metabolic interactions. The model permits the simulation of the carcinogenic estrogen quinones, whose transient nature prevents their direct quantitation. Using rate constants of genetic variants of CYP1A1, CYP1B1, and COMT, the model allows examination of the kinetic effect of enzyme polymorphisms on the entire pathway, including the identification of those haplotypes producing the largest amounts of catechols and quinones. We conceptually addressed the ambiguity of phenotypic information about enzyme concentration by varying the CYP1B1/CYP1A1 ratio. Application of the model to a breast cancer case-control population defined E₂-3,4-Q AUC as a potential risk factor. The model identified a subset of women with an increased risk of breast cancer based on their enzyme haplotype and consequent E₂-3,4-Q production. The model offers for the first time the opportunity to combine genetic, metabolic, and lifetime exposure data in assessing estrogens as breast cancer risk factor.