An Interaction Model for Estimating In Vitro Estrogenic and Androgenic

Department of Radiation Oncology, Temple University, 3401 N. Broad Street, Philadelphia, Pennsylvania 19140, United States. Environ. Sci. Technol. , 0...
2 downloads 0 Views 347KB Size
Article pubs.acs.org/est

An Interaction Model for Estimating In Vitro Estrogenic and Androgenic Activity of Chemical Mixtures Candice M. Johnson,† Mohan Achary,‡ and Rominder P. Suri*,† †

NSF Water & Environmental Technology (WET) Center, Department of Civil and Environmental Engineering, Temple University, Philadelphia, Pennsylvania 19122, United States ‡ Department of Radiation Oncology, Temple University, 3401 N. Broad Street, Philadelphia, Pennsylvania 19140, United States ABSTRACT: There is a need to better understand and predict the biological activity and interaction of chemical constituents in mixtures. Many existing methods assume that the mixture components are additive, and in the case of endocrine disruption, deviation from additivity may occur and render predictions inconclusive. In this study, an alternate index, aRP, which enables the quantification of an antagonistic interaction from analytically derived concentrations of chemical constituents within a mixture that act upon the same molecular target is described. The index is calculated by measuring the degree to which the test compound modulates the activity of a standard hormone as a function of mixture proportions. The aRP was shown to be valid for additive mixtures. It theoretically estimates the product of the relative potential and the interaction index inverse for nonadditive mixtures. The aRP values were computed for agonists and antagonists of both the estrogen and androgen receptors by using yeast-based methods (YES and YAS). The resulting aRP estimates were then validated using higher order mixtures of agonists and antagonists. The use of aRP led to improved predictions compared to estimates based on the toxicity equivalent factor (TEF) approach. The aRP model yielded estimates that were statistically indistinguishable (α = 0.01) from the measured responses in 75% of the 32 mixtures tested. By the same criteria, the TEF approach successfully predicted 34% of the mixtures. Both the aRP and TEF approach correlated well with the observed responses (Pearson R = 0.98 and 0.84, respectively); however, the TEF estimates produced higher percent errors, particularly in mixtures with higher proportions of antagonists. It is suggested that the use of the aRP index allows for a better approximation of the net activity captured by the bioassays through the use of chemically derived concentrations.



INTRODUCTION Environmental hormones of both natural and synthetic origin are widespread in the environment.1 These hormones in combination with other micropollutants can result in complex mixtures that exhibit undesirable effects on aquatic organisms.2 Risk assessors are thus faced with the challenge of determining the permittable amounts of endogenous and exogenous hormones in matrices where analytical capabilities preclude full characterization and determination of total mixture effects.3,4 As such, bioassays have been employed to analyze whole sample effects without having a priori knowledge of the complete sample characterization. The toxicity identification evaluation (TIE) and effect directed analysis (EDA) schemes have been used to identify the natural hormones (17β-estradiol and estrone) in addition to 17α-ethinylestradiol as large contributors to endocrine disruption caused by sewage effluents; whereas, in waters with a heavy industrial influence, nonylphenol becomes important.5−7 It has also been recognized that in addition to estrogenic activation by environmental estrogens, interferences to the activity of the androgen receptor by natural or anthropogenic contaminants can also result in endocrine disruption in vivo.8 It is not only the stimulation of these © 2013 American Chemical Society

nuclear receptors, but also suppression of agonistic activity, which contributes to the observed disruptive effects. Both environmental estrogens and antiandrogens may occur in the same complex mixtures and effect directed approaches have been applied in efforts to identify the most influential estrogenic and androgenic contaminants.9,10 However, the fact that environmental estrogens may act as antiandrogens may complicate the analysis. An essential part of the effect directed approach is the confirmation step. This step confirms that the chemicals identified as causing the effect can sufficiently explain the effect observed in the whole sample.11,12 While there are established methods for confirmation, such as add back and spiking methods,11 the fact that environmental contaminants can display either agonistic or antagonistic hormonal activity may render the results of the confirmation step difficult to interpret, and (more importantly) contaminants resulting in significant (anti-) hormonal activity could be overlooked. A common Received: Revised: Accepted: Published: 4661

April 11, 2012 March 18, 2013 March 20, 2013 March 20, 2013 dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669

Environmental Science & Technology

Article

behavior in mixtures, particularly in cases where the mixture contains chemicals that do not induce an effect alone but are interactive in combination. This study describes the development of a predictive model that can be used to estimate biological activity of mixtures comprising both agonists and antagonists of hormonal activity. It is shown that the relative potentials (used to convert chemical exposure concentrations to the biological responses associated with them) can be calculated through a combined approach in which the test substance is combined with the standard allowing for both agonistic and antagonistic activity to be discerned. The relative potential (RP) of a substance, which possesses antihormonal activity, can therefore be quantified relative to the same standard used for agonists. Importantly, this RP is corrected for interactions between chemicals by including an interaction parameter leading to a new parameter called the alternative relative potency, aRP. The use of aRP allows for the net inductive effect, as measured in bioassays, to be approximated through analytical means. In this contribution, quantification of aRP values is theoretically developed and tested using simulated mixtures in the Yeast Estrogen and Androgen Screening Assays (YES, YAS). The aRP estimates were validated in higher order mixtures of agonists and antagonists of hormonal activity.

approach in confirmation is to compare the induction equivalent concentrations (IEQ), as determined with bioassays for the whole sample, with the analytically determined concentrations of the identified toxicants multiplied by the relative potentials of the toxicants; that is, using the toxicity equivalent factor approach. Agreement between the two values implies that the observed responses in the whole sample can be well explained by the identified causative agents.11 Disagreement may imply strong matrix effects in analytical methods, as a significant amount of sample enrichment may be required. At the same time, disagreemnt may also be due to unidentified compounds, which by themselves produce little or no biological effects, but in combination may alter the activity of the identified agonists and display antihormonal activity.4,13 This antihormonal activity has been observed particularly in androgen screening assays.9,10 An interaction model that can predict antagonistic activity is therefore useful in confirming the biological activity of the whole mixtures from the analytically determined concentrations of both agonists and antagonists. Such a model may not be limited to the determination of total estrogenic and androgenic activity but also has the potential to determine the joint toxicity of chemicals due to mechanisms other than estrogenicity and androgenicity. Several methods exist that statistically test for deviations from additivity;14−20 however, it is less common that models are developed to predict the joint toxicity of multicomponent mixtures containing interactive components. Nonetheless, interaction models have been previously proposed.21,22 Mu et al.21 used an independent joint action model corrected for interactions through the use of k-functions. The independent action model assumes that the probabilities of an organism responding to a specified concentration of each of the constituents of a mixture are statistically independent.23 Rider at al.,22 further developed the approach of Mu et al.,21 by combining the concepts of concentration addition, response addition, and k-functions to assess the toxicity of chemical mixtures. Through the use of the k-functions, the synergistic effects in binary mixtures and the joint toxic effect of acetylcholinesterase and P450 inhibitors were predicted for ternary mixtures. The model was used with some success (r2 = 0.716) and correction with k-values improved estimates of biological activity compared to additive models. Rider and Mu21,22 have shown that k-functions could be used to predict the effective concentration of the mixture, ECmix, from logistic regression equations. In this contribution, the use of k-functions is further extended to facilitate the derivation of induction equivalent concentrations in terms of an agonistic standard. This approach is similar to the toxicity equivalent factor method; however, the method presented appears to have more tolerance for deviations from additivity. Prior studies have used binary mixture toxicity results to predict the toxicity of higher order mixtures with greater accuracy than the concentration addition model.24,25 Results from these studies show that a deviation parameter defined as the sum of an interaction parameter times the relative proportion of toxic units (ci/ECxi; where ci is the concentration of chemical i in a mixture, and ECxi is the effect concentration of chemical i alone that produces the same effect, x%) of each chemical in a mixture could be used to predict the joint toxicity of mixtures. Similarly in this study, the joint mixture effects of ternary and quaternary mixtures are predicted from binary subsets of the mixture components. Further, the model presented could be useful for the prediction of nonadditive



MATERIALS AND METHODS Estriol, 17α-dihydroequilin, bisphenol A, dibutyl phthalate, ethanol (HPLC-grade), and chlorophenolred-β-D-galactopyranoside (CRPG) were obtained from Sigma Aldrich. All other chemicals were obtained from Fisher Scientific. Yeast Based Screening Assays. The Yeast Estrogen Screen (YES) and Yeast Androgen Screen (YAS) were used to quantify the estrogenic and androgenic activity of the standalone components and mixtures. The assays make use of yeast (Saccharomyces cerevisiae) that are genetically transformed with the human Estrogen Receptor (ER) or Androgen Receptor (AR), Estrogen or Androgen Responsive Elements (ERE/ARE), and a Lac Z reporting plasmid. Upon activation of the receptor, β-galactosidase is secreted into the assay medium. The level of β-galactosidase secretion is concentration dependent and relates to the estrogenicity or androgenicity of the sample tested.26 The assay was conducted as per the Routledge and Sumpter method.26 Briefly, stock solutions of 17β-estradiol, testosterone, estriol, 17α-dihydroequilin, dibutyl phthalate, and bisphenol A were dissolved in ethanol. Serial dilutions of each constituent were performed in ethanol, and 5 μL of each constituent was transferred to a flat-bottom 96-well plate and allowed to dry to completeness. Standards, 17β-estradiol (YES), and testosterone (YAS) were used as positive controls on each plate in the respective assay. In each mixture analyzed, the standards were first dried on the plate followed by the addition of the test compound, which was also allowed to dry. Assay medium, which consisted of growth medium, yeast at a final optical density of 0.1, and CPRG, was added upon drying. The samples were then shaken for 2 min and incubated at 32 °C for 72 h for the YES. In the YAS assay, the plates were incubated for 48 h as the color development is faster in YAS assay than in YES.9 The conversion of CPRG to CPR was then quantified using a Biosystems Analyst AD microplate reader at 530 nm, and corrected for turbidity at 630 nm. The concentration−effect relation is absorption of 530 nm light in the in vitro system containing a sample corrected for absorption of the control 4662

dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669

Environmental Science & Technology

Article

Now, an expression for the overall interaction index, γX, can be written as follows:

system (no sample applied) and corrected for the same at 630 nm. The equation for the corrected absorbance is given by eq 1.

n

Corrected Absorbance

γX =

= Sample530 − Control530 − (Sample630 − Control630)

n=1

(1)

(Top − Bottom) 1 + 10(log EC50 − b)H

(2)

where Top and Bottom define the maximum and basal (constrained to zero) responses, respectively, H is the hillslope, EC50 is the 50% effective concentration, and b is the logarithm of the concentration of the compound. The RP estimate is mathematically defined as the ratio of the standard and sample at equi-effective concentrations and yields a unitless estimate of potency (eq 3). As such, RPX =

ECX , S ECX , T

n

(3)

CS +

where the RP of compound T is relative to a standard, S, and ECX is the effective concentration at the X% effect level.

n=1

Cn = γX ECX , n

(4)

where Cn is the concentration of compound n in the mixture exerting an effect of X%, ECX,n represents the effect concentration of n that alone causes an X% effect, and γX represents the interaction index.16,27 If the compounds do not interact with each other but act merely as dilutions of each other, then the mixture is additive and γX is equal to unity. Deviations from additivity result in γX having a value greater than 1 for antagonistic mixtures, or γX less than 1 in cases of synergism.15,17 A suitable predictive equation could be derived from eq 4 as follows. Let compound 1 represent the concentration of the assay standard against which the potency of each mixture constituent will be assessed, denoted from here forth as “S”. Then multiplying both sides of the equation by ECX,S, dividing through by the interaction index and substituting eq 3 yields

aRP = γ −1·RP

n

CS + γX−1 ∑ (RPX , n ·Cn) = ECX , S n=1

(7)

where IEQ represents the induction equivalent concentration of the standard. Note that in the case of additive mixture components, the interaction index is unity and eq 7 represents the toxicity equivalent factor approach. It is therefore suggested that a suitable nonadditive or interactive equation should take into account the inverse of the interaction index of each of the mixture constituents as well as the relative potential. This approach is similar to the toxicity equivalent factor (TEF) approach and in fact, could be considered an extension of the TEF method. In the TEF method, the RP is used as a scaling factor for additive mixture constituents. Equation 7 shows that the RP times the interaction inverse could be used as an appropriate scaling factor for interactive mixture constituents. As such, nonadditive behavior could be accounted for through the incorporation of the interaction index inverse. Although the equation appears straightforward, the traditional approach of calculating the RP (eq 3) does not allow for extended use of this equation. One of the more challenging issues relates to the fact that interactive compounds that result in either antagonism or synergism may not induce an effect when administered alone.9,18 In such cases it is not possible to calculate a relative potential, as there is no measurable induction by the chemical and eq 3 cannot be used. Further, the value of the interaction index is usually not readily accessible data, although it can be derived.18 The development of a novel approach that approximates the γ−1 RP (aRP) is presented in this paper. The aRP metric will ideally be defined by eq 8 such that for additive mixtures (γ = 1) the CA derived TEF equation is maintained.

RESULTS AND DISCUSSION Derivation of the Predictive Equation for an Interaction. The concentration additive model or toxicity equivalent factor method are related to the isobole equation and typically used to predict the concentration−responses of mixtures for which the single-component responses are known.27 Typically, concentration additivity applies when chemicals exert an effect by the same mechanism of action, and the concentration of one compound can be substituted by an equipotent portion of the other. The following isobole equation represents a general description of any multicomponent mixture: n

∑ (γX−,1n·RPX ,n ·Cn) = ECX ,S = IEQ n=1





(6)

where γX,n is the interaction index of compound n in the mixture. It is hyopothesized that the interactions which take place in binary mixtures will be sufficiently strong enough to predict the responses of higher order mixtures. This approach has already been demonstrated by Cedergreen et al.24 It is further hypothesized that the binary interactions between the standard and the test compounds will be strong enough to provide reasonable estimates of higher order mixtures. The total mixture effect could then be evaluated as binary mixtures of the standard and test compounds, particularly if the other test compounds (compound 2 and n for example) do not interact with or alter the absorptive ability of each other. Therefore, the interaction index of any compound other than the nth compound would approximate zero (e.g., γx,2 = 0) for a binary mixture of standard and compound n (γx,n ≠ 0). As such, eq 6 could be represented as eq 7.

The corrected response was then plotted versus the logarithm of the concentration, and the data was fitted to a four-parameter logistic regression concentration response model (version 5.0, GraphPad Software, San Diego, CA) according to eq 2. Y = Bottom +

∑ γX ,n

(8)

Computation of aRP begins with a consideration of the kfunction. The k-function is a multiplier of the concentration of the standard compound in a mix causing some effect, that is

(5) 4663

dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669

Environmental Science & Technology

Article

Figure 1. Modulations observed with combinations of 17β-estradiol with 17α-dihydroequilin (A), and estriol (B). The ratios represent the concentration of standard to test sample. K-function plots for dihydroequilin and estriol are shown in (C) and (D) respectively. K-value represents the ratio of the equal effective concentrations of standard alone to that of the modulated curves (eq 9), and R is the concentration ratio of test compound to standard.

corrected by computing (kX/γX)(d(γX(R))/dR). If (dkR/dR) is zero, then the test compound does not exert an effect and is totally inert with respect to the standard. In this case the compound does not have an effect in the mixture and its equivalent concentration according to eq 7 is zero. As such, the inactivity of the compound is appropriately accounted for in the model. Equation 7 could then be rewritten as eq 14.

needed to obtain the same effect as the standard compound itself. The k-function is mathematically defined as: kX(R ) =

ECX , S ECS(R )

(9)

where ECX,S is the concentration of the standard alone that causes an X% effect; and ECS is the concentration of standard in a mixture causing the same effect.21,22 ECS varies as a function of R, which is the ratio of test compound to standard in a binary mixture of the same. The derivative of the k-function with respect to R can be derived from eq 10 with the standard and test compound concentrations denoted as CS(R), and CT(R) respectively. Multiplying both sides of the equation with the k-function, kX(R) and inserting the definition of the relative potency of the test compound T with respect to standard S, yields eq 10. CS(R ) C (R ) + T = γX(R ) ECX , S ECX , T

(10)

1 + R ·RPX = γX(R ) ·kX(R )

(11)

n

CS +

n=1

⎞ ·Cn⎟ = IEQ ⎠ dR

(14)

The approach appears useful as it takes into account the ability of the test compound to result in an interaction and includes this as a correction factor, γ−1, in the RP estimate. Therefore, in mixtures in which a known interacting agent is present, the induction equivalent (IEQ) estimates can be corrected for the interaction by eq 7, where Cn may either be interactive or additive. The aRP takes into account the interaction and, by extension, the potency change that is expected from that compound in a mixture. Thus, chemical predictions of biological data are expected to better approximate the net biological activity resulting from bioassays. aRP Estimates for Additive Compounds. An investigation into the modulatory activity of the test compound on the standard is the first step in deriving the aRP index. Figure 1 shows the modulatory activity of 17α-dihydroequilin (A) and estriol (B) on 17β-estradiol in the YES assay. In each mixture, a fixed ratio design was employed, in which the proportion of mixture constituents remained constant in the mixture as the mixture was diluted for concentration−response curve formulation. Further, the observed responses were plotted against the concentration of the 17β-estradiol standard, thus the modulating activity could be discerned. The response of the test compound is represented in terms of the standard. It is important to mention that the modulations do not necessarily

In this case, dkX k d(γ (R )) = γX −1·RPX − X · X γX dR dR

⎛ dKR

∑ ⎜⎝

(12)

From eq 12, it is shown that (dkR/dR) approximates aRP, γ−1RP. The second term reflects changes that may occur as the proportion of each mixture constituent changes. A linear relationship between the k-values and R implies that γR is constant over the range of concentrations tested and as such the second term cancels. In this case, (dkR/dR) gives a good approximation of aRP. However, when the k-function plots are not linear, (dkR/dR) yields an approximation of aRP that can be 4664

dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669

Environmental Science & Technology

Article

contaminants than antiestrogens.9,10 Further, several environmental estrogens appear to act as antiandrogens.9 Thus, the presence of antiandrogenic activity in environmental samples may be due to the presence of estrogens, which may affect predictions of androgenic activity. Figure 2 shows the inhibitory activity of dibutyl phthalate (DBP) in the YAS and YES assays, and bisphenol A (BPA) in the YAS assay. None of these compounds induced an effect when tested alone and as such their contribution in a mixture would go unrecognized if the concentration addition-based TEF approach is used to predict activity. The activity of compounds not causing the effect of the standard, but only modulating the standard’s effect are not accounted for in the toxicity equivalent factor approach and as such the estimates for complex samples that may contain these compounds are often inaccurate. The derivative of the kfunction estimates γ−1RP (aRP). This can be used to improve the toxicity equivalent factor approach by scaling the concentration of the compound by γ−1RP to compensate for the activity of such compounds. To obtain the ECS(R) needed for the computation of kvalues, the concentration response curves in Figure 2 were constrained to the maximum response of the respective standards and fitted to the logistic regression equation using eq 2. Since full concentration responses were not obtained for all the BPA modulated curves (due to the toxicity of BPA), the EC15 was used for the derivation of k-functions. Plots of the kvalues versus the ratio of sample to testosterone, R, are shown in Figure 2. Good linearity was observed over the concentration range tested for DBP (r2 = 0.998 and 0.999 in the YAS and YES, respectively); however, the graph was not linear in the case of BPA. The aRP value for BPA varied across the concentration range. Therefore, the data were fitted to a nonlinear equation of the form y = ae−kx and evaluated using the first derivative at the R of interest, i.e. the aRP was estimated as ((dkX/dR)|R). The aRP values for both DBP and BPA in the YAS are shown in Table 2. The aRP for DBP was on the same order of magnitude in the YES and YAS, −6.74 × 10−5 and −5.37 × 10−5, respectively. The aRP values for BPA were given by the

imply that the overall mixture is becoming more potent with increasing concentrations of the test compound, but rather less 17β-estradiol was required to produce the effect observed, as the test compound began to contribute more to the mixture effect.28 To determine the aRP from the modulated curves, the contribution of the test compound to the binary mixture must be quantified. This can be achieved by quantifying the degree to which the test compound modulates 17β-estradiol. This is unique to each compound and relates directly to the relative potential of the test compound. The ratio of the EC50 of 17βestradiol to the EC50 of the modulated curves, i.e. the kfunction, (eq 9), provides a parameter through which the strength of the modulations could be quantified. It measures the reduction in the amount of standard that is required to produce the response at that effect level. The change in the kfunction with respect to the proportion of test sample in the binary mixture (aRP) is constant and linearly related along different mixture proportions as shown in Figure 1. In this linear case, the regression of the k-function versus the ratio of test compound to 17β-estradiol returns a slope (aRP) that is numerically equivalent to the RP of the test compound. Table 1 Table 1. Comparison of Values for aRP and RP Estimates for Additive Compounds in the YES Assay aRP R2 95% confidence intervals RP 95% confidence intervals

17α-dihydroequilin

estriol

0.05094 0.989 0.04359−0.05828 0.04633 0.03909−0.05357

0.006436 0.995 0.006290−0.006582 0.005791 0.005236−0.006346

compares the RP estimates for estriol and 17α-dihydroequilin using both eq 3 and the aRP approach presented. There is an overlap in the 95% confidence intervals of the aRP and RP estimates indicating that they agreed well. This indicates that these compounds act additively toward the estrogen receptor, and validates the method of deriving the aRP estimates since aRP is equivalent to RP for additive mixtures. aRP Estimates for Antagonistic Compounds. The antihormonal assays were conducted using the fixed ratio design as previously described for agonists. This approach differs from the antihormonal screen commonly employed in yeast assays which keeps a submaximal concentration of agonist constant, and varies the amount of the suspected antagonist.9 In that approach the response suppression observed is specific to the amount of standard and antagonist examined. Further, it is common to determine the antihormonal activity in terms of a classic response suppressor, example 4-hydroxytamoxifen (4OHT) in the YES, and flutamide in the YAS leading to the derivation of 4-OHT or flutamide equivalents.10 While these potencies are valid, their use in quantifying the contribution of the inhibitory activity relative to an agonistic standard giving total 17β-estradiol or testosterone equivalents is limited. In this regard, the fixed ratio design is more advantageous as it allows for the representation of the inhibitory activity relative to the same standard to which agonistic activity is compared and the computation of aRP. These aRP values can then be used in a manner similar to RP for the prediction of whole mixture effects through the TEF approach. The YAS assay was used in the antihormonal assays as reports indicate that there are much more antiandrogenic

−4R

equation (dkR/dR) = −3.3 × 10−4 × e2.7×10 . The negative aRP values indicate that the test compounds are suppressing the activity of the standard such that it requires larger amounts of standard to achieve a particular effect with just standard alone. It is intuitive that in a mix containing antagonists, the total induction equivalent concentration represents the contributions of the additive fractions and the negative contributions from antagonists. The negative aRP values facilitate this intuitive notion. As discussed previously, the existing approach to calculating the RP (eq 3) is not suitable for computing the potency of a compound that is inactive when administered alone but modulates the activity of inducers in a mixture. However, such compounds are unarguably potent in the mix and an alternative method to calculating γ−1RP becomes necessary. The derivative of the k-function allows for the computation of the γ−1RP term that is needed to establish the induction equivalent concentrations of such compounds. It is shown that for additive mixtures an evaluation of the derivative of the k-function is numerically equivalent to the RP. This suggests that the RP could be calculated using the new method presented. In the following section, this method is applied to antagonistic compounds DBP and BPA that do not induce an effect when administered alone. 4665

dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669

Environmental Science & Technology

Article

Figure 2. Modulations observed with combinations of testosterone with (A) dibutyl phthalate (DBP) and (B) bisphenol A (BPA) in the YAS assay, and (C) dibutyl phthalate (DBP) in the YES assay. The ratios represent the concentration of standard to the test compounds. K-function plots showing the antihormonal activity of (A) dibutyl phthalate (DBP), and (B) bisphenol A in the YAS, and (C) dibutyl phthalate (DBP) in the YES assay. K-value represents the ratio of the equal effective concentrations of standard alone to that of the modulated curves (eq 9), and R is the concentration ratio of test compound to standard.

Table 2. Comparison of aRP and TEF Predictions to Observed Testosterone Equivalent (TEQ) Concentrationsa aRPDBP −5.37 −5.37 −5.37 −5.37 −5.37 −5.37 −5.37 −5.37 −5.37 −5.37 −5.37 −5.37

× × × × × × × × × × × ×

10−05 10−05 10−05 10−05 10−05 10−05 10−05 10−05 10−05 10−05 10−05 10−05

aRPBPA −2.61 −2.58 −2.55 −2.50 −2.40 −2.22 −1.88 −1.62 −1.36 −1.01 −7.13 −3.91

× × × × × × × × × × × ×

10−04 10−04 10−04 10−04 10−04 10−04 10−04 10−04 10−04 10−04 10−05 10−05

RDBPb

RBPAc

0 5000 5000 5000 5000 5000 5000 10000 5000 10000 5000 10000

0 40 80 160 320 640 1280 1875 2560 3750 5120 7500

TEQ (μg L−1) TEF approach (% error) 2 2 2 2 2 2 2 2 2 2 2 2

(0.1) (39) (35) (46) (53)d (50)d (53)d (433)d (96)d (748)d (133)d (534)d

TEQ (μg L−1) interaction model (% error)

TEQ (μg L−1) observed

2.000 (0.1) 1.437 (0.4) 1.411(5.1) 1.362 (0.3) 1.270 (3.0) 1.109 (17) 0.869 (33) 0.186 (50)d 0.625 (39) 0.039 (83)d 0.630 (27) 0.289 (8)

1.999 1.443 1.487 1.367 1.310 1.333 1.305 0.375 1.024 0.236 0.858 0.417

a Each mixture contained 2 μgL−1 of testosterone. TEQ = Testosterone equivalents. bConcentration ratio of DBP to testosterone. cConcentration ratio of BPA to testosterone. dp < 0.01 (These predictions are significantly different from the observed values).

antagonistic contribution of other phthalates and PAHs, which are known to antagonize the androgen receptor, may result in deviations from the value predicted by additive models. While the objective of this study was to demonstrate the aRP approach, there may exist environmental antagonists that are of higher potency than those examined in this study and which could have greater influence on the activity of the hormone receptors. It is also worth mentioning that the aRP reflects the

There have been previous reports of the antihormonal activity of DBP10,29 and BPA.9 The aRP approach enables the contribution of the antagonistic activity of DBP and BPA on the androgen and estrogen receptor to be quantified in terms of the activity of an agonist standard. The aRP values for DBP and BPA are relatively low, indicating that the ratio of either DBP and BPA needs to be very high in order to have an antagonistic effect in environmental matrices. However, the mixed 4666

dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669

Environmental Science & Technology

Article

Table 3. Comparison of aRP and TEF Predictions to Observed 17β-Estradiol Equivalents (EEQ) Concentrations 17β-E2 (μg L−1)

E3 (μg L−1)

17α- EQN (μg L−1)

DBP (μg L−1)a

EEQ (μg L−1) TEF approach (% error)

EEQ (μg L−1) interaction model (% error)

EEQ (μg L−1) observed

0.0625 0.0625 0.0625 0.0625 0.0625 0.0313 0.0313 0.0313 0.0313 0.0313 0.0313 0.0313 0.0313 0.0313 0.0313 0 0 0 0 0

6.25 6.25 6.25 6.25 6.25 6.25 6.25 6.25 6.25 6.25 6.25 3.125 1.563 0.781 0.391 6.25 6.25 6.25 0 0

0.5 0.5 0.5 0.5 0.5 0.5 0.25 0.125 0.0625 0.0313 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0 4 4

1200 600 300 150 75 1200 600 300 150 75 600 600 600 600 600 600 300 300 1200 600

0.1219 (123)b 0.1219 (59)b 0.1219 (34) 0.1219 (9) 0.1219 (6) 0.0907(136)b 0.0791 (92)b 0.0733 (67)b 0.0703 (64)b 0.0689 (48) 0.0907 (50) 0.0726 (51)b 0.0635 (44b) 0.0590 (42)b 0.0567 (54)b 0.0594 (77)b 0.0594 (50)b 0.0362 (68)b 0.1858 (38) 0.1858 (27)

0.0473 (13) 0.0878 (14) 0.1080 (18) 0.1181 (6) 0.1231 (5) 0.0161 (58)b 0.0438 (6) 0.0576 (31)b 0.0645 (50)b 0.0680 (46) 0.0565 (6.8) 0.0364 (24) 0.0263 (40)b 0.0213 (49)b 0.0188 (49)b 0.0253 (25) 0.0455 (15) 0.0200 (7) 0.1229 (9) 0.1633 (12)

0.0546 0.0765 0.0913 0.1118 0.1292 0.0384 0.0413 0.0439 0.0430 0.0467 0.0607 0.0481 0.0441 0.0416 0.0368 0.0335 0.0396 0.0216 0.1349 0.1459

aRP for DBP is −6.74 × 10−5, EEQ = 17β-estradiol equivalents at the 50% effect level. bp < 0.01 (These predictions are significantly different from the observed values). a

(kX/γX)(d(γX(R))/dR) may be small in comparison to γ−1RP in eq 12. Common factors, such as lack of parallelism, between the curves may lead to some error in both the aRP and TEF appraoch. However, in spite of these errors the aRP model presents an increase in predictive power over the TEF approach as the percent error in only 2 of the aRP predictions exceeded 50%, compared to 17 of the predictions using the TEF approach. The modeled aRP data also closely followed the theoretical 1:1 ideal case scenario, with a slope of 1.085, r2 = 0.961, compared to the TEF with a slope of 0.545, r2 = 0.706, Figure 3.

notion that not only the in vivo concentrations, but also the proportions of contaminants in relation to the natural hormone levels lead to altered receptor activity and potential endocrine disrupting effects. As such, certain stages of development, which are sensitive to disturbances in a delicate hormone balance, such as fetuses and the growing young, may be particularly at risk. Validation of the aRP Estimates. To validate the model, ternary mixtures of DBP, BPA, and testosterone were prepared in the ratios shown in Table 2. Similarly quaternary mixtures of 17β-estradiol, estriol, 17α-dihydroequilin, and DBP were prepared as specified in Table 3. Equation 14 was used to predict the total testosterone or 17β-estradiol equivalents (IEQ) using the aRP values which were previously computed and the results are shown in Tables 2 and 3. Estimates of the total induction equivalents were also made using the TEF approach through the use of eq 7 with γ =1 for additivity. For compounds that did not induce an effect alone (DBP and BPA) the relative potential was set to zero in the additive model. The estimates based on aRP approximated the total IEQs well and showed a significant improvement over the predictions based on the TEF appraoch. The TEF approach tended to overestimate the IEQs given that DBP and BPA did not induce an effect by themselves; and moreover, their interactive nature is only observed when in combination. In 24 out of the 32 mixtures studied, there was no significant difference between the aRP predictions and the observed effect (p > 0.01). Only 11 of the mixtures produced comparable predictions using the TEF approach. When the proportion of antagonist is high, the percent difference between the TEF approach and the observed responses could be as high as >500%. In such mixtures, the aRP model provided better estimates, with the highest percent error being 83%. Although the (kX/γX)(d(γX(R))/dR) term was included with mixtures of BPA, the aRP predictions were still superior to the TEF approach. The good agreement between the aRP model and the observed responses indicate that the

Figure 3. Comparison of observed induction equivalents to predictions based on toxicity equivalent factor (TEF) and the aRP interaction model. The data in Tables 2 and 3 were used to create this figure. The chemicals analyzed included various combinations of estrogens (17β-estradiol, estriol, and 17α-dihydroequilin and dibutyl phthalate), and testosterone with varying amounts of dibutyl phthalate and bisphenol A. 4667

dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669

Environmental Science & Technology

Article

(7) Sheahan, D. A.; Brighty, G. C.; Daniel, M.; Kirby, S. J.; Hurst, M. R.; Kennedy, J.; Morris, S.; Routledge, E. J.; Sumpter, J. P.; Waldock, M. J. Estrogenic activity measured in a sewage treatment works treating industrial inputs containing high concentrations of alkylphenolic compounds - A case study. Environ. Toxicol. Chem. 2002, 21 (3), 507−514. (8) Kelce, W. R.; Wilson, E. M. Environmental antiandrogens: Developmental effects, molecular mechanisms, and clinical implications. J. Mol. Med.-JMM 1997, 75 (3), 198−207. (9) Sohoni, P.; Sumpter, J. P. Several environmental oestrogens are also anti-androgens. J. Endocrinol. 1998, 158 (3), 327−339. (10) Weiss, J. M.; Hamers, T.; Thomas, K. V.; van der Linden, S.; Leonards, P. E. G.; Lamoree, M. H. Masking effect of anti-androgens on androgenic activity in European river sediment unveiled by effectdirected analysis. Anal. Bioanal. Chem. 2009, 394 (5), 1385−1397. (11) Brack, W.; Schmitt-Jansen, M.; Machala, M.; Brix, R.; Barcelo, D.; Schymanski, E.; Streck, G.; Schulze, T. How to confirm identified toxicants in effect-directed analysis. Anal. Bioanal. Chem. 2008, 390 (8), 1959−1973. (12) Grote, M.; Brack, W.; Walter, H. A.; Altenburger, R. Confirmation of cause-effect relationships using effect-directed analysis for complex environmental samples. Environ. Toxicol. Chem. 2005, 24 (6), 1420−1427. (13) Heisterkamp, I.; Gandrass, J.; Ruck, W. Bioassay-directed chemical analysis utilizing LC-MS: A tool for identifying estrogenic compounds in water samples? Anal. Bioanal. Chem. 2004, 378 (3), 709−715. (14) Sorensen, H.; Cedergreen, N.; Skovgaard, I. M.; Streibig, J. C. An isobole-based statistical model and test for synergism/antagonism in binary mixture toxicity experiments. Environ. Ecol. Stat. 2007, 14 (4), 383−397. (15) Tallarida, R. J.; Stone, D. J.; Raffa, R. B. Efficient designs for studying synergistic drug combinations. Life Sci. 1997, 61 (26), PL417−PL425. (16) Tallarida, R. J. Drug Synergism and Dose-Effect Data Analysis; Chapman & Hall/CRC Press: Boca Raton, FL, 2000; p viii, 247 p. (17) Tallarida, R. J. Drug synergism: Its detection and applications. J. Pharmacol. Exp. Ther. 2001, 298 (3), 865−872. (18) Tallarida, R. J. The interaction index: A measure of drug synergism. Pain 2002, 98 (1−2), 163−168. (19) Wu, M.; Woolson, R. F. A comparison of several tests for drug synergy. Commun. Stat.-Simul. Comput. 1998, 27 (2), 303−327. (20) Barrera, N. P.; Morales, B.; Torres, S.; Villalon, M. Principles: Mechanisms and modeling of synergism in cellular responses. Trends Pharmacol. Sci. 2005, 26 (10), 526−532. (21) Mu, X. Y.; LeBlanc, G. A. Synergistic interaction of endocrinedisrupting chemicals: Model development using an ecdysone receptor antagonist and a hormone synthesis inhibitor. Environ. Toxicol. Chem. 2004, 23 (4), 1085−1091. (22) Rider, C. V.; LeBlanc, G. A. An integrated addition and interaction model for assessing toxicity of chemical mixtures. Toxicol. Sci. 2005, 87 (2), 520−528. (23) Backhaus, T.; Arrhenius, A.; Blanck, H. Toxicity of a mixture of dissimilarly acting substances to natural algal communities: Predictive power and limitations of independent action and concentration addition. Environ. Sci. Technol. 2004, 38 (23), 6363−6370. (24) Cedergreen, N.; Sorensen, H.; Svendsen, C. Can the joint effect of ternary mixtures be predicted from binary mixture toxicity results? Sci. Total Environ. 2012, 427, 229−237. (25) Jonker, M. J.; Svendsen, C.; Bedaux, J. J. M.; Bongers, M.; Kammenga, J. E. Significance testing of synergistic/antagonistic, dose level-dependent, or dose ratio-dependent effects in mixture doseresponse analysis. Environ. Toxicol. Chem. 2005, 24 (10), 2701−2713. (26) Routledge, E. J.; Sumpter, J. P. Estrogenic activity of surfactants and some of their degradation products assessed using a recombinant yeast screen. Environ. Toxicol. Chem. 1996, 15 (3), 241−248. (27) Loewe, S. The Problem of Synergism and Antagonism of Combined Drugs. Arzneim.-Forsch., Prog. Drug Res. 1953, 3 (6), 285− 290.

The proposed aRP index appears to be useful in potency balance relationships as the interaction potential is captured and accounted for. It is suggested that aRPX = (dkX/dR) ≅ γ−1 XR RPX,T could be determined as to apply it as a scaling factor for the concentration of a test compound (mixture constituent), T, and this could be used in the future risk assessment of complex mixtures. In cases where the γX(R) is constant as a function of R then the aRP value is independent of the binary mixture composition. This case is presented in the validation shown in Table 3, where mixtures containing inducers of the estrogenic effect, and the antagonist dibutyl phthalate but lacking the standard were tested. The aRP prediction agreed well with the observed responses in these cases. Where γX(R) is not constant, it is hypothesized that the aRP can be determined from the binary mixture at the ratio, R, of the test compound to the total inductive equivalent concentration of the additive mixture constituents in the multicomponent mixture under investigation. While we were not able to identify any such mixtures for our validation experiments this would be an interesting topic for further investigation.



AUTHOR INFORMATION

Corresponding Author

*Phone: (215) 204-2376; fax: (215) 204-0622 rsuri@temple. edu. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS



REFERENCES

This research was partly supported by the National Science Foundation (NSF) I/UCRC Water and Environmental Technology (WET) Center and Temple University. The opinions expressed in this article are those of the authors and do not necessarily reflect the views of the WET Center, NSF or Temple University. We also would like to extend thanks to Dr. Benoit Van Aken and Dr. Mohammad F. Kiani for use of their lab facilities.

(1) Chimchirian, R. F.; Suri, R. P. S.; Fu, H. X. Free synthetic and natural estrogen hormones in influent and effluent of three municipal wastewater treatment plants. Water Environ. Res. 2007, 79 (9), 969− 974. (2) Soto, A. M.; Sonnenschein, C. Environmental causes of cancer: Endocrine disruptors as carcinogens. Nat. Rev. Endocrinol. 2010, 6 (7), 364−371. (3) Roda, A.; Mirasoli, M.; Michelini, E.; Magliulo, M.; Simoni, P.; Guardigli, M.; Curini, R.; Sergi, M.; Marino, A. Analytical approach for monitoring endocrine-disrupting compounds in urban waste water treatment plants. Anal. Bioanal. Chem. 2006, 385 (4), 742−752. (4) Thorpe, K. L.; Gross-Sorokin, M.; Johnson, I.; Brighty, G.; Tyler, C. R. An assessment of the model of concentration addition for predicting the estrogenic activity of chemical mixtures in wastewater treatment works effluents. Environ. Health Perspect. 2006, 114, 90−97. (5) Desbrow, C.; Routledge, E. J.; Brighty, G. C.; Sumpter, J. P.; Waldock, M. Identification of estrogenic chemicals in STW effluent. 1. Chemical fractionation and in vitro biological screening. Environ. Sci. Technol. 1998, 32 (11), 1549−1558. (6) Rodgers-Gray, T. P.; Jobling, S.; Morris, S.; Kelly, C.; Kirby, S.; Janbakhsh, A.; Harries, J. E.; Waldock, M. J.; Sumpter, J. P.; Tyler, C. R. Long-term temporal changes in the estrogenic composition of treated sewage effluent and its biological effects on fish. Environ. Sci. Technol. 2000, 34 (8), 1521−1528. 4668

dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669

Environmental Science & Technology

Article

(28) Rajapakse, N.; Ong, D.; Kortenkamp, A. Defining the impact of weakly estrogenic chemicals on the action of steroidal estrogens. Toxicol. Sci. 2001, 60 (2), 296−304. (29) Okubo, T.; Suzuki, T.; Yokoyama, Y.; Kano, K.; Kano, I. Estimation of estrogenic and anti-estrogenic activities of some phthalate diesters and monoesters by MCF-7 cell proliferation assay in vitro. Biol. Pharm. Bull. 2003, 26 (8), 1219−1224.

4669

dx.doi.org/10.1021/es304939c | Environ. Sci. Technol. 2013, 47, 4661−4669