Mechanistic Computational Model of Steroidogenesis in H295R Cells: Role of Oxysterols and Cell Proliferation to Improve Predictability of Biochemical Response to Endocrine Active Chemical-Metyrapone

Miyuki Breen, *** Michael S. Breen,¿ Natsuko Terasaki,§ Makoto Yamazaki,§ Alun L. Lloyd,t and Rory B. Conolly*,1

*Integrated Systems Toxicology Division, National Health and Environmental Effects Research Laboratory, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina 27711; }Biomathematics Graduate Program, Department of Mathematics, North Carolina State University, Raleigh, North Carolina 27695; ¿ Human Exposure and Atmospheric Sciences Division, National Exposure Research Laboratory, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina 27711; and §Safety Research Laboratory, Mitsubishi Tanabe Pharma Corporation, Kisarazu, Chiba 292-0818, Japan

1To whom correspondence should be addressed at Office of Research and Development, National Health and Environmental Effects Research Laboratory, U.S. Environmental Protection Agency, 109 T. W. Alexander Drive, Mail B105-03, Research Triangle Park, NC 27711. Fax: (919) 541-0694. E-mail: conolly.rory@epa.gov

Received December 28, 2010; accepted June 14, 2011

The human adrenocortical carcinoma cell line H295R is being used as an in vitro steroidogenesis screening assay to assess the impact of endocrine active chemicals (EACs) capable of altering steroid biosynthesis. To enhance the interpretation and quantita- tive application of measurement data in risk assessments, we are developing a mechanistic computational model of adrenal steroidogenesis in H295R cells to predict the synthesis of steroids from cholesterol (CHOL) and their biochemical response to EACs. We previously developed a deterministic model that describes the biosynthetic pathways for the conversion of CHOL to steroids and the kinetics for enzyme inhibition by the EAC, metyrapone (MET). In this study, we extended our dynamic model by (1) including a cell proliferation model supported by additional experiments and (2) adding a pathway for the biosynthesis of oxysterols (OXY), which are endogenous products of CHOL not linked to steroidogenesis. The cell proliferation model predictions closely matched the time-course measurements of the number of viable H295R cells. The extended steroidogenesis model estimates closely correspond to the measured time-course concentrations of CHOL and 14 adrenal steroids both in the cells and in the medium and the calculated time-course concentrations of OXY from control and MET-exposed cells. Our study demonstrates the improvement of the extended, more biologically realistic model to predict CHOL and steroid concentrations in H295R cells and medium and their dynamic biochemical response to the EAC, MET. This mechanistic modeling capability could help define mechanisms of action for poorly characterized chemicals for predictive risk assessments.

Key Words: endocrine disrupting chemicals; mechanistic computational model; in vitro toxicology; metyrapone; H295R cells; steroid biosynthesis.

endocrine disrupting properties of chemicals in drinking water and pesticides. This legislation addresses the concern that various environmental chemical contaminants may alter the endocrine system of humans and wildlife with subsequent adverse outcomes and disease. Based on this legislation, an endocrine disruptor screening program (EDSP) was developed by the U.S. Environmental Protection Agency (U.S. EPA, 1998). The EDSP is designed as a two-tiered screening and testing process to identify chemicals that can interact with the endocrine system (Tier 1) and to characterize their dose- response (Tier 2). In the EDSP Tier 1 battery of screening assays (U.S. EPA, 2009), the human adrenocortical carcinoma cell line H295R was selected as an in vitro steroidogenesis assay to detect chemicals that affect steroid biosynthesis.

As a screening tool, the in vitro H295R steroidogenesis assay has several strengths. This assay can be used to evaluate the effects across the entire steroidogenesis pathway because H295R cells express all the key enzymes for steroidogenesis and have the ability to produce all the adrenocortico steroids as well as sex steroids (Gazdar et al., 1990; Rainey et al., 1994; Staels et al., 1993). The screen is rapid and inexpensive and can detect chemicals that either inhibit or induce steroidogenesis (U.S. EPA, 2009). Furthermore, this assay coupled with a mechanistic computational model supports the recommendations by the National Research Council (NRC) report regarding a vision and strategy for toxicology testing in the 21st century (NRC, 2007). The NRC report recommends the use of in vitro systems to assess mechanisms of action for a large number of chemicals while reducing the number of animals and testing cost for in vivo assays (Andersen and Krewski, 2010).

To increase the understanding and quantitative use of data from the in vitro H295R steroidogenesis assay for human

and ecological risk assessments, we are developing a mech- anistic computational model of steroidogenesis in H295R cells. We previously developed a deterministic model that describes the biosynthetic pathways for conversion of cholesterol (CHOL) to adrenocortical steroids and the kinetics for enzyme inhibition by the competitive steroido- genic enzyme inhibitor, metyrapone (MET), a model endocrine active chemical (EAC) with a well-characterized mechanism of action (Breen et al., 2010). Metyrapone is used as a clinical diagnostic test for pituitary adrenocorti- cotropic dysfunction. In the present paper, we extended the model and performed additional critical experiments to address key limitations of the previously described model.

The primary focus of the previous work was on steroid synthesis (Breen et al., 2010). The main limitation of the previous steroidogenesis model was the large underestimation and overestimation of CHOL concentrations in cells and medium, respectively. In the present paper, we addressed this limitation by investigating CHOL utilization. Because CHOL metabolism is responsible for the biosynthesis of steroids and oxysterols (OXY) (Bjorkhem, 2002; Javitt, 2008; Nishimura et al., 2005; Schroepfer, 2000), we extended the previous model by adding the pathway for OXY biosynthesis. We examined the hypothesis that metabolism of CHOL into OXY may reduce the large discrepancy between measured and model-predicted concentrations of CHOL in cells and medium.

The OXY, similar to steroid hormones, are endogenous products of CHOL. The OXY are formed in many tissues, including adrenal tissue, through CHOL oxygenation reactions mediated by different cytochrome P450 enzymes or reactive oxygen species (Adams et al., 2004; Bjorkhem, 2002; Schroepfer, 2000). The OXY possess potent regulatory functions in a broad range of biological mechanisms, including CHOL homeostasis, apoptosis, calcium uptake, and cell differentiation (Bjorkhem, 2002; Javitt, 2008; Nishimura et al., 2005; Schroepfer, 2000).

Another key limitation of the previous steroidogenesis model is the potential confounding effects of cell pro- liferation and viability on the time-course concentrations of the steroids. Because the proliferation of H295R cells can be substantial (Logie et al., 1999), we addressed this limitation by performing additional experiments and developing a cell proliferation model. Cell proliferation and viability experi- ments for control and EAC-exposed H295R cells were performed, and the data were used for parameter estimation and model evaluation. The cell proliferation model was then linked with the steroidogenesis model to control for this confounding.

The contribution of this study is the extension of a previously developed steroidogenesis model (Breen et al., 2010) by including (1) a pathway for OXY biosynthesis and (2) cell proliferation model. The extended model was evaluated with measurements of CHOL and 14 steroids in H295R cells and medium, measurements from an H295R cell proliferation and

viability assay, and OXY concentrations determined from a molecular balance formulation.

MATERIALS AND METHODS

We first describe the H295R cell proliferation and steroidogenesis experiments and the calculation of OXY concentrations. Then, we present the mathematical models for cell proliferation, pathways for OXY and steroid biosynthesis, and procedures for parameter estimation and sensitivity analysis.

H295R cell proliferation assay. We performed a cell proliferation study following the experimental method as the previously described H295R steroidogenesis assay for control and two concentrations of MET (1 and 10uM) (Breen et al., 2010). The MET concentrations were selected based on cytotoxicity and clinical data. Using a cell viability assay, no significant cytotoxicity was observed at these two concentrations after treatment for 3 days. Clinical observations show mean plasma peak concentrations of 2.2 and 16.3uM at 4 and 1 h following administration of 750 mg MET, respectively (www.pharma.us.novartis.com).

For the cell proliferation assay, 6 × 105 cells were initially incubated for 72 h (prestimuli incubation period). At poststimuli incubation periods of 0, 24, 48, and 72 h, all cells were separated and removed from six replicate wells. The number and percentage of viable cells in each well were then determined using a cell analyzer (Vi-CELL XR; Beckman Coulter, Fullerton, CA).

For statistical analysis of the cell viability data, two-way ANOVA test was used to determine differences between the mean numbers of viable cells across treatments (MET doses), using sampling times as a blocking factor to control variability due to differences across sampling time. Differences were considered significant at p ≤ 0.05.

Steroidogenesis assay with H295R cells. We performed in vitro experi- mental studies with H295R cells: a control study with samples collected at five time points (0, 8, 24, 48, and 72 h) and a MET study with two MET concentrations (1 and 10uM) with samples analyzed at four time points (8, 24, 48, and 72 h). The start time of the experiments was after changing the medium, adding stimuli for activation of steroidogenesis, and including the concentrations of MET. The details were previously described (Breen et al., 2010). Briefly, the medium and cells were separately removed from four replicate wells at each time point. The cells were sonicated in 100 ul of distilled water to produce a cell lysate, which included all the cellular membranes. The CHOL concentrations in the medium and cell lysate were measured using a commercial kit (Wako Pure Chemical Industries, Ltd, Osaka, Japan) based on a CHOL oxidase method. Steroid concentrations in the medium and cell lysate were measured using liquid chromatography/mass spectrometry for 12 steroids (pregnenolone [PREG], 17a-hydroxy-pregnenolone [HPREG], dehydroe- piandrosterone [DHEA], progesterone [PROG], 17a-hydroxy-progesterone [HPROG], androstenedione [DIONE], testosterone [T], deoxycorticosterone [DCORTICO], corticosterone [CORTICO], aldosterone [ALDO], 11-deoxycortisol [DCORT], and cortisol [CORT]) and ELISA for two additional steroids (estrone [E1] and 17ß-estradiol [E2]). The steroid concentrations were adjusted for the recovery of each steroid, with a recovery range between 81.7 and 94.1%. The quantitative ranges for CHOL and each steroid in cells and medium are provided in Supplementary Table S1.

Oxysterols calculated from measurements of cholesterol and steroids. Because no OXY measurements are available, all OXY molecules were lumped together, and the OXY concentrations were calculated based on a molecular balance formulation. To determine OXY concentrations, we made four assumptions. First, we assumed no degradation of CHOL, OXY, and steroids. This assumption is supported by other studies reporting little or no degradation of various steroids across 72 h, our experimental duration (Evans et al., 2001; Garde and Hansen, 2005; Wickings and Nieschlag, 1976). Second, we assumed no de novo synthesis of CHOL because CHOL is abundant in our experiments and is unlikely to be synthesized. Third, because OXY data are unavailable, we

assumed no OXY transport between cells and medium. Fourth, we assumed that the sum of number of CHOL, OXY, and steroid molecules is conserved across time. We determined OXY concentrations from the molecular balance equation, which equates the quantity of molecules (i.e., CHOL, OXY, and steroids) at the initial time to those at later times, as described by

Vi Cdi

cell CHOL,cell + C

d,i OXY,cell +

14 x,cell + V

x=1

d,i=1 CHOL,cell +

d,i=1 OXY,cell +

14 E x=1

C

d,i=1 x,cell

+V

med

C C CHOL,med + d,i=1

14

2

C

d,i=1

x,med

;

x=1

) (1)

where CCHOL,cell and CCHOL,med are the measured concentrations of CHOL in cells and medium at the ith time and dth MET dose (including control) for d = 1, 2, 3 (0, 1, 10uM) and i = 1, … , 5 (0, 8, 24, 48, 72 h), respectively; Cx,cell and Cx,med are the measured concentrations of steroid x in cells and medium at the ith time and dth MET dose, respectively; COXY.cell are the calculated concentrations of OXY in cells at the ith time and dth MET dose; Vien is the volume of cells at the ith time, as predicted by the cell proliferation model described below; and Vmed is the volume of medium. To solve Equation (1) for C di di=‘1 COXY.cell, the initial OXY concentrations in cells are assumed to be zero: COXY,cell = 0. We calculated OXY for one experiment because medium concentrations of El and E2 at time equals zero were below blank sample

d,i concentrations in replicate experiments. We obtained the equation for COXY,cell as

d,i ‘OXY,cell =

+ c Vmed cell

d,i=1 CHOL,cell C

d,i=1 CHOL,med d,i CHOL,cell

d,i CHOL,med +

+

2 x=1 14 C

x,med D

d,i=1 14 x=1

C

)

d,i x,med : ) (2)

Mathematical model of cell proliferation. To predict the number of viable H295R cells per well across time, Ncell(t), we developed a cell proliferation model. We assumed an exponential cell population growth as described by

Ncel(t) = Ncell(-72)ekp(++72), , (3)

where Ncell(-72) is the initial number of viable cells and kp is the proliferation rate. The least squares method was used to estimate kp with time-course data from the control experiments. By taking logs of Equation (3), we obtained a least squares estimate for kp as

K$= TI] TX, (4)

where kp is the least squares estimate of kp, X is a 25-element vector, and T’ = [0 72 96 120 144 … 72 96 120 144]. The vector X is defined as

X = 0

ln Ncell (t = 48,r=1) Ncell(-72)

ln Ncel(t=0,r=1) Ncell (-72) Ncel(-72) ) ln Ncell(t =72,r=1)

Ncel (t =24,r=1)

ln Ncell (-72)

)

)

ln Ncell (t =72,r=6) Ncel(-72)

)]: ?

for t = 0, 24, 48, 72, and r = 1, 2, … , 6, where Ncell (t, r) is the measured number of viable cells at time t and replicate r. Negative time denotes time before stimuli added to initiate steroid biosynthesis (prestimuli). Positive time denotes incubation time with stimuli (poststimuli).

The cell proliferation model was used to estimate the volume of viable cells. From Equation (3), we determined the volume of viable cells per well across time, Vcell(t), as

Vcell(t) = Vcell ( - 72)exp(+72), (6)

where Vcell(-72) is the initial volume of viable cells per well. The Vcell(-72) is obtained by multiplying Ncell(-72) by the overall mean volume of an individual H295R cell from control and the two MET concentrations (1 and 10uM), which was previously determined to be 1499 um3 (Breen et al., 2010). The overall mean volume was used since the mean volumes changed only slightly between controls and the two MET concentrations.

The Vcell(t) was used to compensate for steroid dilution in the cell lysate. We determined the concentration of steroid x in cells, Ccell.x(t), by

Cell,x(+) = Clysatex (1) ( Vysate (*). Vcell (t) ? (7) ;

where Clysate,«(t) is the measured concentration of steroid x in the cell lysate at time t and is the volume of cell lysate at time t. The Vlysate(t) is the sum of Vcell(t) and the volume of distilled water. This Vcell(t) was also used in the dynamic molecular balance equations to determine the time-course concentrations of CHOL, OXY, and the 14 steroids.

Overview of mathematical model for metabolic and transport pathways. The computational model is based on the experimental design with two compart- ments: culture medium and H295R cells (Fig. 1). The model includes two distinct metabolic pathways originating from CHOL: OXY biosynthesis and steroid biosynthesis. The OXY biosynthesis pathway includes conversion of CHOL into OXY. The steroid biosynthesis pathway includes the conversion of CHOL into 14 steroids (PREG, HPREG, DHEA, PROG, HPROG, DIONE, T, DCORTICO, CORTICO, ALDO, DCORT, CORT, E1, and E2) and the inhibition of the steroidogenic enzymes by MET.

The transport pathways for the model include cellular uptake of CHOL and MET and import and secretion of the steroids. The transport of OXY between the cells and medium is not included in the model because no data are available. The details and supporting literature data for the biological pathways of steroid biosynthesis and transport were previously described (Breen et al., 2010). Below, we first describe the details of the metabolic and transport pathways. Then, the dynamic molecular balance equations, which couple the metabolic and transport pathways, are described. The complete set of equations is provided in the Supplementary data.

Metabolic pathways. In the extended model, the conversion of CHOL into OXY is described by a first-order rate equation (Fig. 1B). The metabolic pathway that converts CHOL into the 14 steroids consists of 17 enzymatic reactions catalyzed by nine different proteins (Fig. 1A) (Payne and Hales, 2004). As described in the previous model (Breen et al., 2010), we assumed that the substrate concentration is much less than the Michaelis constant (substrate concentration that yields a half- maximal reaction rate). Thus, the rate of product formation increases linearly with substrate concentration as described by a first-order rate equation (Fig. 1B).

Various EACs can inhibit the enzymes in the steroidogenesis metabolic pathway. In this study, we examined the response of H295R cells exposed to the EAC, MET, which is a competitive inhibitor of cytochrome P450 11ß- hydroxylase (CYP11B1) (Harvey and Everett, 2003; Harvey et al., 2007). For the two CYP11B1 enzymatic reactions competitively inhibited by MET: conversion of DCORTICO to CORTICO and conversion of DCORT to CORT (Fig. 1A), the kinetic parameters k16 and k17 are divided by &CORTICO(t) = 1 + (CMET,cell(t)/k41) and @CORT(t) =1 +(CMET,cell(t)/k42), respectively, where k41 and k42 are MET inhibition constants (Fig. 1B).

Transport pathways. The model used for the transport of CHOL and the steroids between cells and medium was previously described (Breen et al., 2010). Briefly, we model the transport rate of CHOL from the medium as a first-order process (Fig. 1B). We model the secretion and uptake rates for each steroid as reversible first-order processes (k+x and k_x for secretion and uptake of steroid x, respectively) (Supplementary Fig. S1).

(5)

14 x=1

med C

d,i CHOL,med +

C x,med

d,i

¼ Vi-

cell

x=1 14 E C d,i=1 x,cell -

- 2 d,i x,cell

x=1 14

C

A

CYP11A

H295R Cells

CHOL

PREG

CYP17H

HPREG

CYP17L

DHEA

OE

3BHSD2

17BHSD1

OXY

PROG

HPROG

DIONE T

CYP21A

CYP19

DCORTICO

DCORT

E1

E2

StAR

CYP11B1

MET

CORTICO

CORT

CYP11B2

ALDO

CHOL

MET

PROG

CORTICO

HPREG

DCORT

DHEA

E1

E2

PREG

DCORTICO

ALDO

HPROG

CORT

DIONE

T

Medium

B

FIG. 1. (A) Graphical representation of the two compartment model (culture medium and cells) for biosynthesis of OXY and steroids in H295R cells, and enzyme inhibition by MET. Cell uptake of CHOL from medium is depicted by the broad gray arrow labeled with the steroidogenic acute regulatory protein (StAR). Irreversible metabolic reaction for biosynthesis of OXY from CHOL is depicted by white arrow labeled with OXY enzymes (OE). Irreversible metabolic reactions for biosynthesis of 14 steroids (PREG, HPREG, DHEA, PROG, HPROG, DIONE, T, E1, E2, DCORTICO, CORTICO, ALDO, DCORT, and CORT) from CHOL are depicted by arrows, with each pattern representing a unique enzyme. Enzymes labeled next to metabolic reactions they catalyze are: CYP450 side-chain-cleavage (CYP11A), CYP450c17a-hydroxylase (CYP17H), CYP450c17,20-lyase (CYP17L), 3ß-hydroxydehydrogenase type 2 (3ßHSD2), 17ß-hydroxydehydrogenase type 1 (17@HSD1), CYP450 aromatase (CYP19), CYP450 21a-hydroxylase (CYP21A), CYP450 11ß-hydroxylase type 1 (CYP11B1), and aldosterone synthase (CYP11B2). Bidirectional thin gray arrows depict reversible steroid transport between medium and cells. The EAC, MET, is shown as enzyme inhibitor of CYP11B1. (B) Graphical representation of model parameters, which consists of first- order rate constants for CHOL uptake into cells, k1, and for each metabolic process, ko, k2-k18. For quasi-equilibrium analysis, equilibrium constants are q19-q32. Partition coefficient for MET is q40- Enzyme inhibition constants for MET are k41 and k42 for CORTICO and CORT pathways, respectively.

K2

H295R Cells

CHOL

PREG

K3

HPREG

DHEA

Ko

K5

K6

K7

Kg

K10

OXY

PROG

Kg

HPROG

DIONE

T

K11

K12

K13

₭14

DCORTICO

K15

DCORT

E1

E2

16

K41

MET

K42

K17

CORTICO

CORT

18

ALDO

K1

919

920

921

922

923

924

940

125

926

927

928

929 9:30 931

932

CHOL

MET

PROG

CORTICO

HPREG

DCORT

DHEA

E1

E2

PREG

DCORTICO

ALDO

HPROG

CORT

DIONE

T

Medium

As described in the previous model (Breen et al., 2010), we assume no degradation of MET, and MET diffuses into the cells and reaches equilibrium with the MET concentration in the medium

CMET,cell(t) = q40CMET,med(t), (8)

where q40 is the equilibrium coefficient, and CMET,cell(t) and CMET,med(t) are the

cell and medium MET concentrations at time t, respectively (Fig. 1B). To calculate CMET,cell(t), we need to account for the Vcell(t) and Vmed. We obtained the equation for CMET,cell(t) by solving the molecular balance equation

Vcell(t)CMET,cell(t) + VmedCMET,med(t) =Vcell(t)CMET,cell(0) +Vmed CMET,med (0), (9)

for CMET,med(t) and substituted into Equation (8) with CMET,cell(0) = 0 to yield

q40 CMET,cell(t) = ( 1 +q40Vcell (t)/Vmed ) CMET,med (0). (10)

Dynamic molecular balances. The time-course of the steroids, CHOL, and OXY are described by dynamic molecular balance equations (Supplemen- tary data). The general dynamic molecular balance equations for the steroids in cells and medium are

d(Vcell(t)Cx,cell(t)) dt = Px,cell(t) - Ux,cell(t) +Ix,cell(t) -Sx,cell(t), (11)

and

dCx,med(t) dt

Vmed x,med (1) = Sxcell (t) - Ixcell(t), (12)

where Cx,med(t) is the concentration of steroid x in medium at time t; Px,cell(t) and Ux,cell(t) are the production and utilization rates of steroid x in cells at time t, respectively; Ix,cell(t) and Sx,cell(t) are the cell import and secretion rates of steroid x at time t, respectively. The first two terms on the right side of Equation (11) represent the net metabolic reaction rate of steroid x and the last two terms represent the net cellular uptake or release rate of steroid x. The time-courses of CHOL and OXY are calculated in the same manner as the time-courses of the steroids.

Quasi-equilibrium analysis. Based on the evidence that steroid transport between the cells and medium is rapid and reversible, as described in the previous model (Breen et al., 2010), we assume the steroid concentrations in the cells and medium are operating near equilibrium. Because the steroids are involved in the larger network consisting of the metabolic pathway of steroidogenesis, this is considered as quasi-equilibrium.

For quasi-equilibrium, the reversible transport rates (k+x and k_x for secretion and import of steroid x, respectively) are assumed to be much greater than the metabolic reaction rates. The concentration of steroid x in the cells and medium rapidly reaches equilibrium to yield

Cx,med (t) _k+x Cx,cell(t) k_x 9x = qx, (13)

where qx is the equilibrium constant of steroid x transport. By solving Equation (13) for Cx,med(t), we obtain an algebraic equation for each steroid in the medium as

Cx,med(t) = qxCx,cell(t). (14)

To determine the dynamic molecular balance equations for the steroids in cells for quasi-equilibrium, we sum the molecules of steroid x in the cells and medium based on Equations (11 and 12) and substitute Equation (14) for Cx,med(t) to yield

d(Vcell(t)Cx,cell(t) +Vmed Cx,med(t))_d[(Vcell(t)+Vmedqx)Cx,cell(t)] dt

= Px,cell(t) - Ux,cell (t). dt

(15)

We obtain a system of equations consisting of a differential equation for each steroid in the cells from Equation (15) as

Downloaded from http://toxsci.oxfordjournals.org/ by guest on June 4, 2015

dCx,cell (t)1Px,cell(t)- Ux,cell(t) -Cx,cell(t)dVcell (t)
dt¼ Vcell(t) + Vmedqxdt ;
(16)
where
dVcell(t) = kp Vcell(-72)ekp(+72) = kpVcell(t). dt(17)

Equation (17) was obtained by differentiating Equation (6) with respect to time. The quasi-equilibrium assumption reduces the number of the model parameters to 36 parameters: 14 transport equilibrium constants (q19, q20, … , q32), 18 metabolic rate constants (k0, k2, k3, … , k18), CHOL import rate (k1), 2 enzyme inhibition constants for MET (k41 and k42), and the partition coefficient for MET (q40). These dynamic molecular balance equations for quasi-equilibrium and the 36 parameters are used in all subsequent analysis (Supplementary data).

Parameter estimation. The parameters for the transport and metabolic pathways were independently estimated using the mean concentrations from replicate experiments. For the transport pathway, the equilibrium constants (q19, q20, … , q32) were estimated with the time-course data from the control and MET studies using the least squares method. From Equation (14), we obtained a least squares estimate for qx as

qx = [Cx,cell’ Cx,cell] Cr,cell’ Cr,med, (18)

where qx is the least squares estimate of the equilibrium constant for steroid x transport, and = cd=1,i=1 ‘x,cell C x,cell . .. ad=1,i=2 cd=3,i=5 x,cell and Cx,med’ = x,med Cx,cell’ cd=1,i=2 d=1,i=1 Cx,med C d=3,i=5] … are the measured concentra- tions in the cells and medium at the ith time and dth MET dose (including x,med control), respectively.

For the metabolic pathway, the parameters (k0, k1, k2, … , k18, k41, k42) were estimated with the time-course data from the control and MET studies. The weighted least squares method was used to estimate these parameters instead of the least squares method to account for CHOL and all the steroids concentrations that vary by several orders of magnitude. Let CCHOL,cell (t; CMET,med, k) and COXY,cell (ti; CMET,med, k ) be the model-predicted concentrations of CHOL and OXY in the cells at the ith time, ti, for the dth MET dose (including control), CMET,med, with parameter set k = (ko, k1, k2, . .. , k18, k41, k42) for d= 1, … , 3, and i = 1, … , 5, respectively; Cx,cell (ti; CMET,med, k) be the model-predicted concentrations of steroid x in the cells at time t; for the MET dose, CMET,med, with parameter set k; CCHOL,cell and COXY,cell be the mean measured concentration of CHOL and calculated concentration of OXY in the cells across time, respectively; Cx.cell be the mean measured concentration of steroid x in the cells across time. Then, the weighted least squares estimate, k” = (kg, k1, k2, … , kjg, k41; k42), is the parameter values k, which minimizes the cost function ☒

J (k) ¼

3 E

1

d=1 CHOL,cell i=1

3 d=1 14 3

C Cd,i x,cell

d,i OXY,cell Cx,cell CCHOL,cell cell

ti; Cd ti; cd

MET,med

+ +

5 1 2 OXY,cell i=1 5 2 x=1d=1 “x,cell i=1 C

1 5

d,i CHOL,cell

COXY,cell k)) ti; Cd *)) 2

Parameters for the metabolic pathway were estimated with a nonlinear optimization algorithm using MATLAB R2010a (Mathworks, Natick, MA) software. The Nelder-Mead simplex method was used due to its relative insensitivity to the initial parameter values as compared with other common methods, such as Newton’s method and its robustness to discontinuities (Nelder and Mead, 1965).

Sensitivity analysis. We performed a sensitivity analysis to examine parameter uncertainty using the method previously described (Breen et al.,

2010). Briefly, the sensitivity function relates changes of the model output to changes in the model parameters. We calculated relative sensitivity functions Rx,med,ki (t) andRx,med,q; (t) with respect to parameters ki and qi, respectively, for each of the model-predicted concentrations in the medium, Cx,med, and each MET dose (including control). MATLAB was used to numerically solve the partial derivatives in Rx,med,k; (t) and Rx,med,q; (t). To rank the relative sensitivities, we calculated the L2 norm across time for each relative sensitivity function as described by

L2 norm(Rx,med,k;)

¼

V 14

Rx,med,k;(t)|2dt (20)

and

L2 norm(Rx,med,;) =|Rx,med,q:(+)2dt. (21)

RESULTS

Cell Proliferation- Estimated Parameter and Model Evaluation

The time-course measurements for the number of viable cells per well (Supplementary Fig. S2A) and percentage of viable cells per well (Supplementary Fig. S2B) are shown for each MET dose. The percentage of viable cells per well across all measurements was 91.8 ± 2.3% (mean ± SD). The mean number of viable cells measured across MET doses (control, 1, and 10uM) was not statistically different (p = 0.41). Therefore, control data were used to estimate the value of kp, which was determined to be 0.00878/h. The model-predicted number of viable cells was compared with time-course measurements from the control experiments (Fig. 2). The model predictions closely correspond to the mean time-course data.

Calculated Oxysterols

FIG. 2. Time-course of measured and model-predicted number of viable cells. Measurements were plotted at 72 h prestimuli and four poststimuli incubation periods of 0, 24, 48, and 72 h (mean ± SD) for control experiments. Negative time denotes time before stimuli added to initiate steroid biosynthesis (prestimuli). Positive time denotes incubation time with stimuli (poststimuli).

3× 106

Model-Predicted

2.5

Measured

Number of Viable Cells

2

1.5

1

0.5

0

-72

-48

-24

0

24

48

72

Time (h)

2 (19)

2

MET,med

MET,med? k)) :

FIG. 3. Left column shows time-course of total (sum in cells and medium) number of measured CHOL and steroid molecules in control (A), 1 uM (C), and 10UM (E) MET-exposed cells. Right column shows time-course for number of calculated OXY molecules for control (B), 1uM (D), and 10uM (F) MET-exposed cells. To calculate OXY concentrations, measured concentrations of CHOL and each steroid are needed. Because medium concentrations of two steroids (E1 and E2) at time t = 0 were considered valid (i.e., above blank sample concentrations) only in one experiment, the single experiment was used to calculate OXY concentrations.

A

B

Number of Molecules per Well (nmoles)

200

Measured CHOL: Control

200

180

Measured Steroids: Control

Number of Molecules per Well (nmoles)

Calculated OXY: Control

180

160

160

140

140

120

120

100

100

80

80

60

60

40

40

20

20

0

0

8

24

Time (h)

48

72

0

0

8

24

Time (h)

48

72

C

D

200

Number of Molecules per Well (nmoles)

Measured CHOL: MET 1AM

200

Calculated OXY: MET 1uM

180

Measured Steroids: MET 1 u.M

Number of Molecules per Well (nmoles)

180

160

160

140

140

120

120

100

100

80

80

60

60

40

40

20

20

0

0

0

8

24

48

72

0

8

24

48

72

Time (h)

Time (h)

Number of Molecules per Well (nmoles) m

F

200

Measured CHOL: MET 10UM

Number of Molecules per Well (nmoles)

200

Calculated OXY: MET 10UM

180

Measured Steroids: MET 10UM

180

160

160

140

140

120

120

100

100

80

80

60

60

40

40

20

20

0

0

0

8

24

48

72

0

8

24

Time (h)

48

72

Time (h)

Figure 3 shows the time-course for the sum of measured number of CHOL and steroid molecules and the calculated number of OXY molecules at each MET dose (control, 1, and 10uM). For each MET dose, the calculated number of OXY molecules monotonically increased across time. The number of OXY molecules showed no apparent pattern with increasing MET dose (Supplementary Fig. S3).

Transport Pathways for Steroids

Table 1 shows the estimated transport equilibrium parameters. As described in the previous model (Breen et al., 2010), the MET transport equilibrium, q40, could not be determined from the data because MET was not measured in the cells. Therefore, q40 was set equal to q22, the CORTICO transport equilibrium, because the partition coefficients for MET

TABLE 1 Estimated Transport Equilibrium Parameters (dimensionless) from Model Fit of Steroids Corresponding to Given q Parameter
ParameterValue
q190.013
q200.005
q210.041
q220.056
q230.091
q240.061
q250.021
q260.042
q270.068
q280.038
q290.040
9300.027
q310.044
q320.035
9400.056ª

ªMetyrapone transport equilibrium (q40) set to CORTICO transport equilibrium (q22); see text for details.

(XLogP = 2.0) and CORTICO (XLogP = 1.9) are similar (PubChem database).

Metabolic Pathways for OXY and Steroid Biosynthesis

Table 2 shows the estimated parameters for the metabolic pathways. The convergence time for the nonlinear parameter estimation was typically around 10 min on an Intel Core 2 Duo processor using MATLAB.

TABLE 2 Estimated Parameters of Metabolic Pathway
ParametersValueUnits
k00.014Per h
0.016Per h
k20.011Per h
0.757Per h
k31.268Per h
k4 k5 k6 k70.814Per h
11.153Per h
7.217 0.177Per h
Per h
kg1.754Per h
kg0.048Per h
k106.479Per h
k1112.188Per h
k120.595Per h
k130.001Per h
k140.091Per h
k150.637Per h
k160.247Per h
k17 k180.122Per h
k4163.566nM
k4225.208nM
FIG. 4. Model evaluation of CHOL and OXY for control and MET-exposed cells. Time-course of model-predicted concentrations were plotted and compared with concentrations (mean ± SD) measured at five sampling times for CHOL in medium (A), CHOL in cells (B), and calculated for OXY in cells (C). To calculate OXY concentrations, measured concentrations of CHOL and each steroid are needed. Because medium concentrations of two steroids (E1 and E2) at time t = 0 were considered valid (i.e., above blank sample concentrations) only in one experiment, the single experiment was used to calculate OXY concentrations.

A

10x 10

Model-Predicted CHOL: Control

Concentration in Medium (nM)

Measured CHOL: Control

8

-Model-Predicted CHOL: MET 1AM

Measured CHOL: MET 1AM

-Model-Predicted CHOL: MET 10UM

. Measured CHOL: MET 10UM

6

4

2

%

10

20

30

40

50

60

70

80

Time (h)

B

4× 10

4

Model-Predicted CHOL: Control

Measured CHOL: Control

Concentration in Cells (nM)

-Model-Predicted CHOL: MET 1uM

3

Measured CHOL: MET 1AM

-Model-Predicted CHOL: MET 10UM

HICH

Measured CHOL: MET 10UM

2

HEOH

%

0

10

20

30

40

50

60

70

80

Time (h)

C

4

x 107

Model-Predicted OXY: Control

Calculated OXY: Control

Concentration in Cells (nM)

-Model-Predicted OXY: MET 1uM

3

Calculated OXY: MET 1AM

-Model-Predicted OXY: MET 10AM

. Calculated OXY: MET 10 UM

2

1

C

0

10

20

30

40

50

60

70

80

Time (h)

For CHOL and OXY, we compared model-predicted concentrations with time-course data from control and MET- exposed cells (Fig. 4). Model-predicted concentrations corre- spond well to the mean time-course data for CHOL both in the

cells and in the medium and OXY in cells. For MET-exposed cells, model-predicted and measured concentrations of CHOL and calculated concentrations of OXY remained approximately unchanged from controls as MET increased.

For CHOL in medium and cells, the extended model performed remarkably better than the previous model (Breen et al., 2010). For CHOL in medium, the extended model overestimated the mean measurements by 9, 13, and 11% at 24, 48, and 72 h, respectively, whereas the previous model overestimated the mean measurements by 43, 96, and 153% at 24, 48, and 72 h, respectively. For CHOL in cells, the extended model overestimated the mean measurements by 2, 4, and 5% at 24, 48, and 72 h, respectively, whereas the previous model underestimated the mean measurements by 52, 53, and 47% at 24, 48, and 72 h, respectively.

For the 14 steroids, we compared model-predicted concentrations with time-course measurements from control and MET-exposed cells. Overall, model-predicted concentrations correspond closely to the mean time-course measurements in cells (Supplementary Fig. S4) and medium (Fig. 5) for control cells. As compared with the previous model (Breen et al., 2010), the extended model better captured the mean time-course behavior for the five steroids (PROG, HPROG, DHEA, HPREG, and PREG), which increased until 48 h and then sharply decreased at 72 h (Figs. 5B, 5C, and 5E). For these five steroids, the extended model predictions increased until 40-50 h and then decreased (Figs. 5B, 5C, and 5E), whereas the previous model predictions monotonically increased across time (Breen et al., 2010). For the other steroids in control cells, the extended model performed similar to the previous model. A detailed evaluation was reported previously (Breen et al., 2010).

For MET-exposed cells, we compared model-predicted steroid concentrations with time-course measurements. For three steroids (ALDO, CORTICO, and CORT) downstream from the enzyme inhibited by MET (CYP11B1), the model- predicted concentrations correspond well to the mean time- course measurements both in the cells (Supplementary Fig. S5) and in the medium (Figs. 6A-C). For two steroids (DCOR- TICO and DCORT) immediately upstream from CYP11B1, the model-predicted concentrations correspond closely to the mean time-course data both in the cells (Supplementary Fig. S5) and in the medium (Figs. 6D and 6E), which remained approxi- mately unchanged at 8, 24, and 48 h as MET increased and slightly increased at 72 h as MET increased. For the other nine steroids further upstream from CYP11B1, model-predicted and measured concentrations remained approximately unchanged from controls as MET increased (data not shown). For five steroids (ALDO, CORTICO, CORT, DCORTICO, and DCORT), the extended and previous models showed similar results for MET-exposed cells.

Sensitivity Analysis

Figures 7 and 8 show the relative sensitivities for the four primary steroids: ALDO, CORT, T, and E2. For ALDO, k18 was highly sensitive at each MET dose, and eight parameters were

moderately sensitive: parameters associated with transport path- ways (k1, q21, and q22), metabolic pathways (k2, k5, and k16), and MET-mediated enzyme inhibition (q40 and k41). For CORT, q27 and k17 were highly sensitive, and two parameters associated with MET-mediated enzyme inhibition (q40 and k42) were moderately sensitive. For steroids T and E2, the parameters associated with MET (q40, k41, and k42) were not sensitive, and the sensitivity of all parameters was unchanged with increasing MET dose. For T, k10 was highly sensitive, and six parameters were moderately sensitive: parameters associated with transport pathways (k1 and q29) and metabolic pathways (k2, k3, k9, and k12). For E2, six parameters were moderately sensitive: parameters associated with transport pathways (q29, q30, and q32) and metabolic pathways (k2, k13, and k15). The E1 pathway appears to be the preferred pathway for E2 synthesis because E2 was sensitive to the E1 pathway (k13 and k15) and not sensitive to the T pathway (k10 and k14). This preferred pathway result is consistent with our previous study of ovarian steroidogenesis (Breen et al., 2007). The sensitivity analysis orders the inputs by importance, identifying main contributors to the variation in the outcome of a model. Parameters with high sensitivity are more important and significant for the model output than parameters with low sensitivity.

DISCUSSION

We extended a computational model of adrenal steroido- genesis to include (1) a cell proliferation model to account for time-varying number of viable cells and (2) a metabolic pathway for biosynthesis of OXY to examine the hypothesis that metabolism of CHOL into OXY improves the model fit for CHOL. Experiments were designed and performed to evaluate the cell proliferation model. The extended model and cell proliferation experiments addressed key limitations of the previous model (Breen et al., 2010) by (1) removing the confounding effects of cell proliferation from the steroidogen- esis model, (2) reducing the large discrepancy between the measured and model-predicted concentrations of CHOL both in the medium and in the cells, and (3) allowing the steroidogenesis model to more accurately capture the observed dynamic behavior. Furthermore, the model’s predictive ability improved considerably with only a slight increase in the model complexity by adding one parameter for cell proliferation and one parameter for OXY biosynthesis.

Metabolic Pathway for Oxysterols

In the previous steroidogenesis model, we developed a steroidogenesis model and evaluated its ability to predict only the steroid concentrations for MET-exposed H295R cells (Breen et al., 2010). The present study was initiated after we discovered that (1) the differences between the model-predicted and measured CHOL concentrations both in the medium and in the cells were large and (2) the sum of the number of measured CHOL and steroids molecules was not conserved across time.

FIG. 5. Model evaluation of steroids for control cells. Time-course of model-predicted concentrations in medium were plotted and compared with concentrations (mean ± SD) measured at five sampling times for ALDO, E2, and T (A); PROG, HPROG, and DHEA (B); HPREG, DIONE, and E1 (C); CORTICO and DCORTICO (D); and PREG, CORT, and DCORT (E).

A

350

-Model-Predicted ALDO

300

Measured ALDO

Concentration in Medium (nM)

Model-Predicted E2

Measured E2

250

Model-Predicted T

Measured T

200

150

100

50

K

0

10

20

30

40

50

60

70

80

Time (h)

B

C

250

-Model-Predicted PROG

1500

Model-Predicted HPREG

Measured PROG

Measured HPREG

Concentration in Medium (nM)

200

Model-Predicted HPROG

Concentration in Medium (nM)

1250

Model-Predicted DIONE

Measured HPROG

Measured DIONE

Model-Predicted DHEA

Model-Predicted E1

Measured DHEA

1000

Measured E1

150

750

100

500

50

250

OM

0

10

20

30

40

50

60

70

80

0

10

20

30

40

50

60

70

80

Time (h)

Time (h)

D

E

8000

-Model-Predicted CORTICO

10000

Model-Predicted PREG

Measured CORTICO

Model-Predicted DCORTICO

Measured PREG

Concentration in Medium (nM)

Model-Predicted CORT

6000

Measured DCORTICO

Concentration in Medium (nM)

7500

Measured CORT

Model-Predicted DCORT

Measured DCORT

4000

5000

2000

2500

C

0

10

20

30

40

50

60

70

80

0

10

20

30

40

50

60

70

80

Time (h)

Time (h)

To explain the lack of a molecular balance in the data, we examined the hypothesis that other metabolic pathways were needed in the model. The literature shows that OXY biosynthesis is a primary pathway for CHOL metabolism, and the pathway is present in adrenal tissue. To test this hypothesis, we (1) included the OXY metabolic pathway in the extended model, (2) calculated OXY concentrations, (3) estimated model parameters, and (4) evaluated the model- predicted concentrations of CHOL, OXY, and all 14 steroids both in the cells and in the medium.

The results support our hypothesis. By including the OXY metabolic pathway, the extended model significantly improved the model fit for CHOL both in the medium and in the cells as compared with the previous model. Moreover, the model-predicted and calculated OXY concentrations closely correspond. Close correspondence of model-predicted and measured CHOL and 14 steroids supports our model assumption of no degradation of CHOL, OXY, and steroids.

Besides the pathway for conversion of CHOL to OXY, we examined alternative biologically realistic hypotheses to allow for

FIG. 6. Model evaluation of steroids for MET-exposed cells. Time-course of model-predicted concentrations in medium were plotted and compared with concentrations (mean ± SD) measured at five sampling times for ALDO (A); CORTICO (B); CORT (C); DCORTICO (D); and DCORT (E). For controls, model- predicted and measured concentrations are same as shown in Figure 5.

A

350

Model-Predicted ALDO: Control

300

Measured ALDO: Control

Concentration in Medium (nM)

Model-Predicted ALDO: MET 1AM

Measured ALDO: MET 1uM

250

-Model-Predicted ALDO: MET 10AM

Measured ALDO: MET 10uM

200

150

100

50

0

10

20

30

40

50

60

70

80

Time (h)

B

C

5000

Model-Predicted CORTICO: Control

3000

Model-Predicted CORT: Control

Measured CORTICO: Control

Measured CORT: Control

Concentration in Medium (nM)

4000

Model-Predicted CORTICO: MET 1 AM

Concentration in Medium (nM)

Model-Predicted CORT: MET 1AM

Measured CORTICO: MET 1uM

-Model-Predicted CORTICO: MET 10UM

2250

Measured CORT: MET 1uM

-Model-Predicted CORT: MET 10UM

Measured CORTICO: MET 10UM

. Measured CORT: MET 10AM

3000

1500

2000

$

750

1000

0

0

10

20

30

40

50

60

70

80

0

10

20

30

40

50

60

70

80

Time (h)

Time (h)

D

E

14000

Model-Predicted DCORTICO: Control

12000

Measured DCORTICO: Control

Model-Predicted DCORT: Control

Model-Predicted DCORTICO: MET 1AM

Measured DCORT: Control

Concentration in Medium (nM)

10000

-Model-Predicted DCORT: MET 1uM

10500

Measured DCORTICO: MET 1uM

Concentration in Medium (nM)

-Model-Predicted DCORTICO: MET 10UM

Measured DCORT: MET 1AM

Measured DCORTICO: MET 10UM

8000

-Model-Predicted DCORT: MET 10AM

Measured DCORT: MET 10UM

7000

6000

4000

3500

2000

0

10

20

30

40

50

60

70

80

10

20

30

40

50

60

70

80

Time (h)

Time (h)

a molecular balance. One alternative hypothesis is an unaccounted form of CHOL because CHOL is distributed in various cell membranes and compartments with a high abundance in the plasma membrane, endocytic recycling compartment, and Golgi complex (Ikonen, 2008). Another alternative hypothesis is a pathway for conversion of CHOL to CHOL esters because CHOL is a biosynthetic precursor of steroid hormones, OXY, and CHOL esters in all cells (Ikonen, 2008). However, in this study, the assay used to measure CHOL concentrations in cells includes CHOL esters and all the cellular membranes and compartments that can contain CHOL and CHOL esters. Because this CHOL

measurement accounts for free, membrane-bound, and esterified CHOL, these alternative hypotheses are not supported.

Dynamic Steroid Behavior

For the previous model, the dynamic steroid predictions for all 14 steroids had the same qualitative behavior, increasing monotonically across time (0-80 h). Although the mean measured concentrations for nine steroids increased mono- tonically, the mean measurements for five steroids (PROG, HPROG, DHEA, HPREG, and PREG) increased until 48 h and then decreased at 72 h. This was a key limitation of the

FIG. 7. Relative sensitivities for model-predicted ALDO (A) and CORT (B) plotted for 36 model parameters (ko-k18, 919-932, q40, k41-k42) for control and MET-exposed cells. Each bar represents L2 norm of relative sensitivities across time (0-80h) and indicates the degree to which changes in parameter values lead to changes in model outputs.

A

30

ALDO

☒ Control

☐ ΜΕΤ 1μ.Μ

25

☐ ΜΕΤ 10μ.Μ

Relative Sensitivity

20

15

10

5

0

1

m

ko

K1

K2

K3

K4

K5

K6

K7

Kg

Kg

k

K1

K1

K1:

K14

K1

K1

K17

K18

919

920

92

922

92

924

925

926

927

928

929

930

931

932

940

K41 1 K42

B

Parameter

30

CORT

☒ Control

☐ ΜΕΤ 1μ.Μ

25

☐ ΜΕΤ 10μ.Μ

Relative Sensitivity

20

15

10

5

0

1

ko

K1

k2

K3

K4

K5

K6

K7

Kg

Kg

K10

K11

K12

K1

K14

K15

K16

K17

K18

919 920

921

922 923

924

925

926

927

928

929

930

931 932

940

K41

K42

Parameter

previous model. By including cell proliferation, the extended model captures this observed time-course behavior for each steroid. One possible factor responsible for this behavior is the increasing number of enzymes due to cell proliferation. This is further evidence that the more biologically realistic features of the extended model allow for a better representation of the observed time-course behavior of steroidogenesis.

Future Applications of Extended Model

There are several potential applications for the extended model. First, the model’s better ability to predict the time-course

of CHOL concentrations both in the medium and in the cells will be critical for EAC that affect upstream metabolic (e.g., inhibition of steroidogenic enzyme CYP11A) or signaling processes, which may affect CHOL levels. Second, the more biologically realistic model may improve the accuracy for low concentration extrapolations of concentration-response curves for other EACs with environmental concentrations below experimental levels. Environmental concentrations of MET are unknown. Third, the model can be simply modified with EAC- specific enzyme inhibition constants to predict the biochemical response for other EACs that competively inhibit steroidogenic

FIG. 8. Relative sensitivities for model-predicted T (A) and E2 (B) plotted for 36 model parameters (ko-k18, q19-q32, 940, k41-k42) for control and MET- exposed cells. Each bar represents L2 norm of relative sensitivities across time (0-80 h) and indicates the degree to which changes in parameter values lead to changes in model outputs.

A

30

T

☒ Control

ΜΕΤ 1μ.Μ

25

☐ ΜΕΤ 10μ.Μ

Relative Sensitivity

20

15

10

5

0

m

m

ko

K1

K2

K3

K4

k5

K6

k7

kg

Kg

K10

K11

1 K12

K13

K14

K1

K16

K17

K18

919

920

921 922 923

924

925

926

927

928

929

930

931

932

940

K41

K42

Parameter

B

30

E2

☒ Control

☐ MET 1µM

25

ΜΕΤ 10μ.Μ

Relative Sensitivity

20

15

10

5

0

m

m

ko

K1

K2

K3

K4

K5

K6

K7

Kg

Kg

K10

K11

K12

K13

K14

15

K16 K17

K18

919 920

921

922

923

924

925

926

927

928

929 930

931

932

940

K41

K42

Parameter

enzymes. Fourth, our in vitro model could be extended to predict the in vivo response. The steroidogenesis model would need to be linked to a multiorgan systems model, which includes the regulatory feedback of the hypothalamus-pituitary-adrenal axis, and refined based on additional experiments. This extension of the current model would require a significant research effort. Finally, with further model refinement and evaluation, the model can be used to help identify mechanisms of action for EACs by predicting the enzyme inhibition constants for poorly character- ized EACs and for screening assays that typically measure only a few steroids in the medium.

Limitations

There are some limitations to our extended model based on the model structure and assumptions and data available for model evaluation. First, the extended model structure and parameter values are based on MET concentrations at or below 10uM. At higher MET concentrations, the proliferation rate and viability of the cells can be altered and inhibition of additional steroidogenic enzymes can occur. Therefore, the extended model may not accurately extrapolate at higher MET concentrations without including a cell proliferation model and enzyme inhibition constants that are dose dependent. Second, our model

assumption of first-order enzyme kinetics, which reduced the model complexity while maintaining the model’s predictive ability, is only applicable for nonsaturable enzyme kinetics. For highly concentrated or potent EACs, first-order enzyme kinetics may need to be replaced by saturable enzyme kinetics (e.g., Michaelis- Menten). Third, transport of OXY between the cells and medium is not included in the model because the OXY data are unavailable. This may result in the overestimation of OXY in the cells. Experiments that measure the time-course of OXY are needed to further evaluate the OXY metabolic and transport pathways. Finally, the extended model structure is based on EACs that are competitive enzyme inhibitors. For EACs with different mecha- nisms of action (e.g., activating or antagonizing steroid hormone receptors and inducing steroidogenic enzymes), model refinements will be needed (Sanderson, 2006). These refinements may require additional model-guided experiments for other pathways, such as gene regulation and upstream signaling pathways.

Conclusions

We extended a previous steroidogenesis model by including a cell proliferation model and a pathway for OXY biosynthesis. The cell proliferation model, which was independently evaluated with experimental data, removed the confounding of cell proliferation from the steroid biosynthesis and its biochemical response to EAC. The inclusion of the metabolic pathway for OXY biosynthesis significantly improved the model fit for CHOL and allowed the model to capture the dynamic behavior of steroids both in the medium and in the cells. Our study demonstrates the significant improvement of the extended, more biologically realistic model to estimate CHOL and adrenal steroid concentrations both in H295R cells and in medium and their dynamic biochemical response to the EAC, MET. This mechanistic modeling capability could help define mechanisms of action for poorly characterized chemicals and mixtures for predictive risk assessments and to screen drug candidates based on steroidogenic effects in the early phase of drug development.

SUPPLEMENTARY DATA

Supplementary data are available online at http://toxsci. oxfordjournals.org/.

FUNDING

North Carolina State University/Environmental Protection Agency Cooperative Training Program in Environmental Sciences Research (Training Agreement CT833235-01-0 to M.B.) with North Carolina State University.

ACKNOWLEDGMENTS

We thank Hisham El-Masri and Daniel Villeneuve for their review comments and helpful suggestions. Although this

manuscript was reviewed by the U.S. EPA and approved for publication, it may not reflect official Agency policy. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.

REFERENCES

Adams, C. M., Reitz, J., De Brabander, J. K., Feramisco, J. D., Li, L., Brown, M. S., and Goldstein, J. L. (2004). Cholesterol and 25-hydroxycholesterol inhibit activation of SREBPs by different mecha- nisms, both involving SCAP and Insigs. J. Biol. Chem. 279, 52772-52780. Andersen, M. E., and Krewski, D. (2010). The vision of toxicity testing in the 21st century: moving from discussion to action. Toxicol. Sci. 117, 17-24. Bjorkhem, I. (2002). Do oxysterols control cholesterol homeostasis? J. Clin. Invest. 110, 725-730.

Breen, M. S., Breen, M., Terasaki, N., Yamazaki, M., and Conolly, R. B. (2010). Computational model of steroidogenesis in human H295R cells to predict biochemical response to endocrine-active chemicals: model de- velopment for metyrapone. Environ. Health Perspect. 118, 265-272.

Breen, M. S., Villeneuve, D. L., Breen, M., Ankly, G. T., and Conolly, R. B. (2007). Mechanistic computational model of ovarian steroidogenesis to predict biochemical responses to endocrine active compounds. Ann. Biomed. Eng. 35, 970-981.

Evans, M. J., Livesey, J. H., Ellis, M. J., and Yandle, T. G. (2001). Effect of anticoagulants and storage temperatures on stability of plasma and serum hormones. Clin. Biochem. 34, 107-112.

Food Quality Protection Act. (1996). Public Law 104-170. U.S. Senate and House of Representatives, Washington, DC.

Garde, A. H., and Hansen, A. M. (2005). Long-term stability of salivary cortisol. Scand. J. Clin. Lab. Investig. 65, 433-436.

Gazdar, A. F., Oie, H. K., Shackleton, C. H., Chen, T. R., Triche, T. J., Myers, C. E., Chrousos, G. P., Brennan, M. F., Stein, C. A., and La Rocca, R. V. (1990). Establishment and characterization of a human adrenocortical carcinoma cell line that expresses multiple pathways of steroid biosynthesis. Cancer Res. 50, 5488-5496.

Harvey, P. W., and Everett, D. J. (2003). The adrenal cortex and steroidogenesis as cellular and molecular targets for toxicity: critical omissions from regulatory endocrine disrupter screening strategies for human health? J. Appl. Toxicol. 23, 81-87.

Harvey, P. W., Everett, D. J., and Springall, C. J. (2007). Adrenal toxicology: a strategy for assessment of functional toxicity to the adrenal cortex and steroidogenesis. J. Appl. Toxicol. 27, 103-115.

Ikonen, E. (2008). Cellular cholesterol trafficking and compartmentalization. Mol. Cell Biol. 9, 125-138.

Javitt, N. B. (2008). Oxysterols: novel biologic roles for the 21st century. Steroids 73, 149-157.

Logie, A., Boulle, N., Gaston, V., Perin, L., Boudou, P., Le Bouc, Y., and Gicquel, C. (1999). Autocrine role of IGF-II in proliferation of human adrenocortical carcinoma NCI H295R cell line. J. Mol. Endocrinol. 23, 23-32.

National Research Council. (2007). Toxicity Testing in the 21st Century: A Vision and a Strategy. National Academy Press, Washington, DC.

Nelder, J. A., and Mead, R. (1965). A simplex method for function minimization. Comput. J. 7, 308-313.

Nishimura, T., Inoue, T., Shibata, N., Sekine, A., Takabe, W., Noguchi, N., and Arai, H. (2005). Inhibition of cholesterol biosynthesis by 25-hydroxycho- lesterol is independent of OSBP. Genes Cells 10, 793-801.

Payne, A. H., and Hales, D. B. (2004). Overview of steroidogenic enzymes in the pathway from cholesterol to active steroid hormones. Endocr. Rev. 25, 947-970.

Rainey, W. E., Bird, I. M., and Mason, J. I. (1994). The NCI-H295R cell line: a pluripotent model for human adrenocortical studies. Mol. Cell. Endocrinol. 100, 45-50.

Safe Drinking Water Act Amendments. (1996). Public Law 104-182. U.S. Senate and House of Representatives, Washington, DC.

Sanderson, J. T. (2006). The steroid hormone biosynthesis pathway as a target for endocrine-disrupting chemicals. Toxicol. Sci. 94, 3-21.

Schroepfer, G. (2000). Oxysterols: modulators of cholesterol metabolism and other processes. Physiol. Rev. 80, 361-554.

Staels, B., Hum, D. W., and Miller, W. L. (1993). Regulation of steroidogenesis in NCI-H295R cells: a cellular model of the human fetal adrenal. Mol. Endocrinol. 7, 423-433.

U.S. Environmental Protection Agency (U.S. EPA). (1998). Endocrine Disruptor Screening Program. Vol. 63 No. 154, pp, 42852- 42855. Federal Register, Washington, DC.

U.S. Environmental Protection Agency (U.S. EPA). (2009). Endocrine Disruptor Screening Program (EDSP): Announcing the Availability of the Tier 1 Screening Battery and Related Test Guidelines. Vol. 74, No. 202, pp. 54415-54422; Federal Register, Washington, DC.

Wickings, E. J., and Nieschlag, E. (1976). Stability of testosterone and androstenedione in blood and plasma samples. Clin. Chim. Acta 71, 439-443.

Downloaded from http://toxsci.oxfordjournals.org/ by guest on June 4, 2015