Mechanism of Inhibition of Estrogen Biosynthesis by Azole Fungicides

Abstract

Biosynthesis of estrogens from androgens is catalyzed by cytochrome P450 aromatase. Aromatase inhibition by the triazole compounds letrozole (LTZ) and anastrozole is a prevalent therapy for estrogen-dependent postmenopausal breast cancer. Azoles are widely used as agricultural fungicides and antimycotic drugs that target 14α-demethylase. Some were previously shown to inhibit aromatase, thereby raising the possibility of endocrine disruptive effects. However, mechanistic analysis of their inhibition has never been undertaken. We have evaluated the inhibitory effects of 3 common fungicides, bifonazole, imazalil, and flusilazole, in human aromatase purified from placenta and compared them with LTZ, the most potent inhibitor of aromatase. Bifonazole exhibits strong inhibitory effects with an IC₅₀ of 270nM and K_i (Michaeles-Menten inhibition constant) of 68nM, compared with 10nM and 13nM, respectively, for LTZ. The IC₅₀ and K_i are 1100nM and 278nM for imazilil and 3200nM and 547nM for flusilazole, respectively. Analyses of inhibition kinetics suggest that the modes of inhibition by azole fungicides are mixed or competitive, whereas LTZ inhibition could be noncompetitive or mixed. We interpret the inhibition mechanism in the context of the x-ray structure of aromatase-androstenedione complex. Structural data show that aromatase has 3 binding pockets in relation to the heme. The substrate-binding cavity at the heme-distal site closely compliments the structures of the natural substrate, androstenedione, and steroidal aromatase inhibitors. Because the structures of LTZ and the azole fungicides are entirely dissimilar to the androstenedione backbone, the azoles possibly inhibit by binding to a structurally rearranged active site, the 2 other catalytically important sites, or both, in agreement with the kinetics data.

Nearly 600 million pounds of triazole and imidazole fungicides are used worldwide in agriculture per year, tens of millions in the United States alone (1–4). Azoles also find their use as antifungal drugs in human mycoses. The primary target of azole compounds is 14α-demethylase (CYP51A1), which catalyzes the early stages of sterol biosynthesis for the formation of functional fungal cell membranes (5, 6). However, azole fungicides (AFs) are known to disrupt normal aromatase function at the transcriptional level affecting the gene expression (7–9). Aromatase (CYP19A1) catalyzes the biosynthesis of estrone, estradiol, and estriol from their respective androgenic precursors androstenedione (ASD), testosterone, and 16α-hydroxy testosterone in the presence of redox partner cytochrome P450 reductase (CPR). Aromatase inhibitors (AIs) exemestane, letrozole (LTZ), and anastrozole are 3 leading drugs in estrogen deprivation therapy for estrogen-dependent postmenopausal breast cancers (10–12). AFs have been shown to inhibit aromatase activity in prepared microsomes (5, 6, 13–15), H295R human adenocarcinoma cells (16), and aromatase-expressing MCF-7 breast cancer cells (17). However, the mechanism of direct inhibition has not been investigated in purified enzyme. Aromatase as an “off” target of AFs could be associated with adverse side effects and form a basis for endocrine disruption (5, 6, 13). Fungicides have been reported to cause testicular tumors, bladder transitional cell tumors, and hepatocellular tumors in rats and mice (18). Exposure of fresh water fish to fungicides leached from soil and plant irrigation systems causes reproductive and developmental side effects (19). Farmers constantly exposed to fungicides have an increased incidence of prostate cancer (20, 21).

The x-ray structure of human placental aromatase has provided the detailed architecture of the active site and the molecular basis of its interaction with substrates and steroidal inhibitors (22–25). Nevertheless, the molecular mechanism of inhibition of aromatase by nonsteroidal LTZ, anastrozole, and AFs is still not understood. Here, we determine and analyze the inhibition kinetics of 3 AFs, the antifungal drug bifonazole (BFZ) and 2 agricultural fungicides, imazalil (IMZ) and flusilazole (FLZ), in purified human placental aromatase and compare them with LTZ. Finally, we describe the inhibition mechanism on the basis of the molecular structure of aromatase.

Materials and Methods

Chemicals

Emulgen913 was purchased from Desert Biologicals, n-dodecyl-β-D-maltopyranoside from Affymetrix, [1β-³H, 4-¹⁴C] ASD from PerkinElmer and LTZ from LKT Laboratories. AFs, 100% polyethylene glycol BioUltra 400 (PEG400) (relative molecular mass: 380–420), and other chemicals were purchased from Sigma-Aldrich.

Purification of aromatase from human placenta

Aromatase was purified from human placenta to homogeneity by immunoaffinity chromatography using a well-established procedure (22, 23), under the Institutional Review Board protocol 2011.1704, Crouse Hospital (Syracuse, NY). For enzyme activity assays, purified aromatase was in buffer A (containing 100mM potassium phosphate buffer [pH 7.4], 20% glycerol, 10μM ASD, 500nM EDTA, and 1mM n-dodecyl-β-D-maltopyranoside).

Monitoring of the Soret peak shift by UV/visible spectrophotometry

The inhibitors were dissolved to 2mM in 100% PEG400 solution and incubated for 16 hours at 4°C with 13μM aromatase in buffer A for an azole/aromatase ratio of 23:1. Soret peak transitions were monitored by scanning the mixture from 250nm to 600nm on an Implen nano-spectrophotometer with a path length of 1.0 mm.

Measurement of aromatase activity

The specific activity of aromatase was determined by the release of tritiated water after the ASD-to-estrone aromatization reaction using reduced nicotinamide adenine dinucleotide phosphate as an electron source (26). All 4 compounds were dissolved in 100% PEG400 solution and serially diluted. The inhibitor working stocks were mixed with 35nM aromatase in buffer A such that the final concentration of PEG400 was 10% (vol/vol), whereas the inhibitor concentrations were as follows: LTZ (500nM, 100nM, 50nM, 10nM, 5nM, and 2.5nM), BFZ (10μM, 5μM, 1μM, 0.5μM, 0.3μM, 0.2 μM, and 0.1μM), IMZ (10μM, 5μM, 1μM, 0.5μM, and 0.25μM), and FLZ (10μM, 5μM, 2.5μM, and 1μM). Any PEG400 concentrations higher than 10% caused protein instability. The samples were incubated at 4°C in the dark for 16 hours. We performed the assay at 2 conditions: one with a fixed concentration of ASD and varied concentration of the inhibitor and the other with a fixed concentration of inhibitor and varied concentration of ASD (from 0nM to 232nM). In the standard control, the substrate was maintained at the saturating concentration of 58nM. The enzyme activity was adjusted for an 11.5% reduction in the presence of 10% PEG400 (the percent reduction ranged from 5% to 15% in 15 independent experiments performed in triplicate). Any inconsistent data were remeasured, and outliers were identified as data 2 SDs from the mean.

Determination of IC₅₀

The potency of each compound was determined by nonlinear regression analysis by GraphPad Prism (27) to generate the best-fit value for log[IC₅₀]. The analysis also generated a SD for log[IC₅₀] that was used to calculate a 95% confidence interval (CI).

Analyses of inhibition kinetics profiles

Regression analyses were performed to generate Michaelis-Menten (M-M) and Lineweaver-Burk (L-B) (data not shown) plots. Statistical analysis, determination of inhibition mode, and calculation of kinetics parameters (Michaelis-Menten Constant [K_M], Michaelis-Menten Inhibition Constant [K_i], and maximal velocity [V_max] followed the standard equations below relating these parameters to the data:

$$K_M^{apparent}=K_M\times \left(1+\frac{\left[I\right]}{K_i}\right)\text{-----------Competitive}$$

$$V_{\max }^{apparent}=\frac{V_{\max }}{\left(1+\frac{\left[I\right]}{K_i}\right)}\text{------------Non}-\text{competitive}$$

$$\begin{array}{l}{K}_M^{apparent}=K_M\times \left(1+\frac{\left[I\right]}{\alpha K_i}\right)\hfill \\ V_{\max }^{apparent}=\frac{V_{\max }}{\left(1+\frac{\left[I\right]}{\alpha K_i}\right)}\hfill \end{array}\}\text{--------Uncompetitive}$$

$$\begin{array}{l}{K}_M^{apparent}=K_M\times \frac{\left(1+\frac{\left[I\right]}{K_i}\right)}{\left(1+\frac{\left[I\right]}{\alpha K_i}\right)}\hfill \\ V_{\max }^{apparent}=\frac{V_{\max }}{\left(1+\frac{\left[I\right]}{\alpha K_i}\right)}\hfill \end{array}\}\text{--------Mixed}$$

V_max, K_M, and K_i were shared among the entire data set so that a single best-fit value could be obtained (27). The mixed model equation includes an additional parameter, α, which determines the mechanism of inhibition (28, 29). When α = 1, the mixed model is identical to noncompetitive inhibition. If α ≫ 1, the mixed-model becomes identical to competitive inhibition. For 1 > α > 0, the mixed model is nearly identical to an uncompetitive model (28, 29). We inquired whether uncompetitive inhibition should be considered as another possible mode and determined that our plots were totally unlike the uncompetitive mode. Calculation of R² values (goodness of fit) agreed with this determination.

Each inhibition model was subjected to a head-to-head comparison with another using the extra sum of squares (SS) F-test (F-test) and Akaike information criterion (AIC) to determine the mode of inhibition. F-test is based on the traditional hypothesis testing paradigm. GraphPad uses a nested model for this test (6, 13, 16), in which one model is simpler and has fewer degrees of freedom (DF) than the other (6, 13, 16). The simpler model (competitive and noncompetitive inhibition) is the null hypothesis (H₀) and the complex model (mixed inhibition) is the alternative hypothesis (H_a). If the simpler model is correct, then the relative increase in the SS results in an increase in DF. An F-ratio is generated for computing P so that H₀ could be accepted or rejected given a threshold P = .05. As shown below, F-ratio is the ratio of SS for the simpler model (SS1) and the more complicated model (SS2).

$$F-ratio=\frac{\left(SS1-SS2\right)/SS2}{\left(DF1-DF2\right)/DF2}$$

When 2 models had the same number of parameters, the P value was calculated manually (6, 13, 16).

AIC is based on information theory and maximum likelihood principle, and the results are expressed in probability that either model is correct (30). It determines how well the data supports each model, taking into account SS1, SS2, and DF values in the equation

$$\Delta AIC=N\times \text{In}\left(\frac{SS2}{SS1}\right)+2\Delta DF$$

for N data points. Model 1 (simpler) or 2 (complicated) is chosen depending on whether ΔAIC is positive or negative, respectively. Simply, the model with smaller AIC is more likely to be correct.

Determination of possible ligand-binding site

The structural organization of human placental aromatase with the N-terminal helix has been previously described (22, 31). MOE2012.10 Site Finder application (32) was used for the determination of ligand binding cavities in aromatase. The method was validated by removing the experimentally observed ASD from the model before binding cavity computation. Q-Site Finder (33) and CASTp Server (34) were used in calculating the volume of each ligand-binding cavity. The calculated active site volume was validated against the previously reported manual calculation (22). The University of California at San Francisco Chimera package (35) was used for visualization and illustration.

Results

Kinetics parameters of aromatase activity

Aromatase from human placenta is purified to homogeneity (Figure 1A). Integrity of the purified enzyme is assessed by Soret absorbance spectra and tritiated water assay (Table 1) (26). The average specific activity of aromatase in the presence of a saturating concentration (58nM) of ASD is 53 ± 1 nmol/min · mg. The M-M constant (K_M) of ASD is 16nM, and V_max is 65 nmol/min · mg (Table 1 and Figure 1B), in agreement with previous reports (36).

Figure 1.

Enzyme kinetics profile of aromatase activity. A, An SDS-PAGE gel showing a typical sample of purified placental aromatase used for measuring kinetic parameters of aromatase activity. B, M-M plot showing the effect of increasing substrate (³H-ASD) concentration on the reaction velocity of aromatase. Data points are averages of 9 separate experiments measured in triplicate. The K_M of ASD is 16nM and the V_max is 65 nmol/min · mg. C–F, M-M plots showing the kinetics profiles of aromatase inhibition by LTZ, BFZ, IMZ, and FLZ. C, The graphical representation of aromatase inhibition by LTZ suggests noncompetitive binding mode. The plots showing aromatase inhibition by (D) BFZ, (E) IMZ, and (F) FLZ suggest a competitive or mixed-type mode of inhibition. ASD concentration ranged from 0nM to 232nM in all cases except for the assay with IMZ (which ranged from 0nM to 174nM). All measurements were conducted in triplicate. V₀, intial velocity; kDa, kilodaltons; 2mer, dimer.

Table 1.

Potency and Kinetics Profile of Placental Aromatase Activity in the Presence and Absence of Inhibitors

	Soret Peak Type	IC50 (nM) (95% CI)	Vmax (nmol/min · mg) (95% CI)	KM of ASD (nM) (95% CI)	Ki (nM) (95% CI)
Aromatase control	I	N/A	65 (63–69)a	16 (14–18)a	N/A
Aromatase-LTZ	II	10 (9.3–10.5)	59 (57–61)	14 (12–16)	13 (6–19)
Aromatase-BFZ	II	270 (230–318)	66 (63–67)	18 (15–20)	84 (68–99)
Aromatase-IMZ	II	1100 (1000–1200)	65 (63–68)	24 (20–27)	278 (232–324)
Aromatase-FLZ	II	3200 (2600–3800)	83 (80–86)b	23 (20–27)b	547 (399–695)

Potency and kinetics of aromatase inhibition by LTZ and AFs as determined by the ³H-ASD assay. IC₅₀s are in the nM range. LTZ is the most potent AI overall. BFZ is the most potent AF.

^aAverage of 9 separate experiments from the same aromatase preparation performed in triplicate. All V_max and K_M values are determined by averaging the individual measurements from each substrate concentration over the range of inhibitor concentrations assayed.

^bAromatase from a different preparation with intrinsically higher specific activity was used for assays with FLZ. V_max and K_M in the absence of FLZ are 90 nmol/min · mg (95% CI: 85–97) and 24nM (95% CI: 17–29), respectively.

Inhibition of aromatase by LTZ, BFZ, IMZ, and FLZ

The summary of inhibitory potencies and kinetics parameters of aromatase inhibition by LTZ and AFs, as assessed by the tritiated water method (26), are shown in Table 1. Like the nonsteroidal AIs, the AFs shift the aromatase Soret peak from type I (high-spin ferric) at 393nm to type II (low-spin ferric) at 420nm (Table 1). Type I peak represents the substrate- or steroidal inhibitor-bound, fully oxidized state of aromatase. Type II peak is the characteristic nonsteroidal AI-bound state (37, 38).

LTZ is the most potent azole overall with an IC₅₀ of 10nM and K_i of 13nM (Table 1). The binding of LTZ to aromatase causes a reduction of V_max to 59 nmol/min · mg, without changing the K_M of ASD (Table 1). Visual/manual examination of M-M (Figure 1C) and L-B (data not shown) plots and the kinetics parameters may suggest noncompetitive inhibition. However, when the error limits of the experimental data are taken into consideration, it is not a strict case of noncompetitive inhibition (V_max decreases with increasing [I] in M-M plots; L-B plots do not all intersect at 1 point on the x-axis). Because other possible modes cannot completely be ruled out, statistical analysis is used to calculate the probability of each inhibition mode (Table 2). In a head-to-head comparison of competitive and noncompetitive modes of aromatase inhibition by LTZ (H₀: competitive), P < .0001 ≪ 0.05, the H₀ is rejected in favor of the H_a. Given H₀: competitive and H_a: mixed, P < .0001, suggesting that the mixed mode is more probable. However, with H₀: noncompetitive and H_a: mixed, P < .1581. Therefore, the noncompetitive model cannot be rejected altogether. With the same initial assumptions, the AIC test predicts with 99.99% probability that the mixed or the noncompetitive models are favored over competitive inhibition, whereas the mixed and noncompetitive are equally probable in their head-to-head analysis (Table 2).

Table 2.

Statistical Analyses of Aromatase Inhibition Kinetics

	Comparison of Fits Using F-Test			Comparison of Fits Using AIC			Conclusion
	C vs NC	C vs M	NC vs M	C vs NC	C vs M	NC vs M
Aromatase-LTZ	P < .0001 (H0: C)	P < .0001 (H0: C)	P < .1581 (H0: NC)	> 99.99% probability that NC is the correct model	> 99.99% probability that M is the correct model	50% probability of NC or M as the correct model	M/NC
Aromatase-BFZ	P < .0001 (H0: NC)	P < .0007 (H0: C)	P < .0001 (H0: NC)	> 99.99% probability that C is the correct model	99.31% probability that M is the correct model	> 99.99% probability that M is the correct model	M
Aromatase-IMZ	P < .0001 (H0: NC)	Simpler model C preferred	Simpler model NC preferred	> 99.99% probability that C is the correct model	Simpler model C preferred	Simpler model NC preferred	C
Aromatase-FLZ	P < .01 (H0: C)	P < .0963 (H0: NC)	P < .0001 (H0: NC)	> 99.99% probability that C is the correct model	58.75% probability that M is the correct model	> 99.99% probability that M is the correct model	M/C

Statistical analyses of the kinetics of aromatase inhibition by LTZ, BFZ, IMZ, and FLZ as determined by F-test and AIC. C, competitive mode inhibition; NC, noncompetitive mode inhibition; M, mixed mode inhibition.

The IC₅₀ and K_i of BFZ are 270nM and 84nM, respectively (Table 1). In the presence of BFZ, V_max, and K_M are virtually unchanged (Table 1), and the M-M plot shows that the mode of inhibition could be mixed or competitive (Figure 1D). The manually calculated P value, assuming that H₀: noncompetitive and H_a: competitive, is less than 0.0001 (Table 2), indicating that competitive is preferred to noncompetitive. Assuming that H_a: mixed and H₀: competitive or noncompetitive, F-test generates P < .0001 in favor of the mixed model. AIC calculates a 99% likelihood of mixed being the correct model over competitive or noncompetitive.

With IMZ as the inhibitor, the IC₅₀ and K_i are 1100nM and 278nM, respectively (Table 1). V_max remains the same, within the limits of error, as that of the uninhibited sample, whereas the K_M of ASD increases to 24nM. Although the M-M plot is suggestive of competitive or mixed inhibition (Figure 1E), F-test and AIC show that the data do not fit the mixed model when compared with simpler models (competitive or noncompetitive). By F-test, P < .0001, the competitive model is favored (H₀: noncompetitive), and AIC also predicts it as the correct model with more than 99.99% probability (Table 2).

FLZ has an IC₅₀ of 3200nM and K_i of 547nM (Table 1). V_max is slightly decreased from 90 to 83 nmol/min · mg, and K_M remains virtually the same (Table 1). There is a modest decrease (<10%) in V_max within the limits of error (Figure 1F). As per AIC, the probability of the mode of inhibition being noncompetitive is less than 0.01% (Table 2). Additionally, by F-test, P < .01 (H₀: noncompetitive), better than the cut-off P < .05. This allows us to preferentially select competitive over noncompetitive mode of inhibition. However, with H₀: competitive and H_a: mixed, P < .096, and hence the preferred model cannot be determined with certainty. AIC confirms nearly equal likelihood of mixed and competitive inhibition (Table 2).

Determination of possible ligand-binding sites

Computation with the aromatase structure reveals 3 ligand-binding cavities in relation to the heme moiety (Figure 2) (22, 25). Site 1, the active site distal to heme, comprises residues that form the catalytic substrate-binding pocket, which is the site of interaction for ASD, other natural androgenic substrates, and steroidal AIs. The access channel, site 2, links the active site to the outer surface facing the lumen of the endoplasmic reticulum, and putatively, facilitates the entry and exit of lipophilic steroids. Site 3 is at the heme-proximal region and comprises of the residues closest to the cysteine thiolate ligand of the heme (24, 25, 31). Higher-order associations among aromatase molecules (24, 39), as well as coupling with CPR, occur at this interface. The volume of the active site, heme-proximal cavity, and access channel are calculated to be approximately 400, approximately 700, and approximately 500 ų, respectively.

Figure 2.

Binding sites in aromatase. The aromatase molecule has 3 binding cavities in relation to the heme moiety. The residues that form the catalytic substrate-binding pocket define site 1, the active site, distal to the heme. It is the site of interaction for the natural androgenic substrates and steroidal AIs. Site 2, the access channel, links site 1 site to the outer surface facing the lumen of the endoplasmic reticulum. It may facilitate entry and exit of lipophilic steroids. Site 3 is the heme-proximal region, which comprises the residues closest to cysteine thiolate ligand of the heme. It is the site for intermolecular association, interaction with CPR, and interaction with small molecules.

Discussion

This work unequivocally establishes that AFs impede estrogen biosynthesis by direct inhibition of aromatase enzyme activity and should thus be regarded as endocrine disrupting chemicals. The azoles studied exhibit high potency and could, therefore, pose threat to human when exposed routinely. The kinetics data and statistical analyses show that all azole compounds exhibit the characteristics of mixed inhibition. Although some have the dominant features of competitive inhibition, others are more noncompetitive in nature.

Aromatase uses one site for catalysis (substrate-binding cavity) and the other for coupling of the electron transfer partner (heme-proximal region). Therefore, classic definitions of competitive and noncompetitive inhibition may not be strictly valid in the context of aromatase inhibition when elements of both sites are involved. The active site closely compliments the structure of natural substrates (22) and steroidal AIs (25). Because the structures of LTZ and the AFs are entirely dissimilar to the ASD backbone, accommodating these compounds in the active site would require substantial rearrangement. Docking of the nonsteroidal AIs to the active site of aromatase causes severe steric clashes (Jiang, W., and D. Ghosh, unpublished results). LTZ and some of the AFs could bind at the active site by inducing large conformational changes and yielding pseudocompetitive kinetics as suggested by our data. Alternately, the binding of the azoles at an away site could allosterically induce conformational changes obstructing the ASD cavity. However, analysis of the normal modes suggests that the steroid-binding core is rigid and has least fluctuations (25).

If an inhibitor binds to the aromatase-ASD complex at the heme-proximal site, it could 1) disrupt the coupling interface and flow of electrons from CPR to aromatase; and/or 2) cause conformational changes at the active site by allostery. On the other hand, any primary or secondary site at or near the access channel would directly block the passage of substrate/product causing enzyme inhibition. Lastly, LTZ and AFs could bind at yet another unknown site and allosterically alter the active site conformation. In all these scenarios, the classical definitions of competitive and noncompetitive inhibitions may not be strictly applicable. The binding modes need to be further confirmed by experimental determination of the crystal structures of aromatase-LTZ/AF complexes.

Direct inhibition of aromatase by AFs could not only disrupt the androgen-estrogen ratio and classic endocrine functions of sex steroids but also interfere with their nongenomic signaling and protective roles in CNS. When administered on fungal infections as prescribed, BFZ has minimum inhibitory dose ranging 200nM–13μM (40, 41). As a broad-spectrum plant fungicide, IMZ has a maximum residue limit of 200 parts per million (42), whereas that of FLZ used primarily for soybeans is 2.6 parts per million (43). FLZ is applied at 448 g/acre, with a half-life of 71–140 days (18). Although no correlation can yet be made between these numbers and the inhibition parameters determined, the question of how much is absorbed by human organs and tissues needs to be answered in earnest. Given the established residue limits and half-life, the inhibitory potencies we have reported suggest that the AFs could have significant effects. Understanding the molecular details of interaction of AFs with aromatase and CYP51A1 would help design better fungicides with minimal disrupting effects on estrogen biosynthesis.