|
|
Experimental assays were carried out at the NCTR for the specific purpose of deriving data to train models to predict binding to the estrogen receptor (ER) and the androgen receptor (AR).
NCTR used a validated (standardized) estrogen receptor (ER) competitive binding assay to determine the ER affinity for a large, structurally diverse group of chemicals [Blair, 2000 #216]. Uteri from ovariectomized Sprague-Dawley rats were the ER source for the competitive binding assay. Initially, test chemicals were screened at high concentrations to determine whether a chemical competed with [3H]-estradiol for the ER. Test chemicals that exhibited affinity for the ER in the first tier/screening were subsequently assayed using a wide range of concentrations to characterize the binding curve and to determine each chemical's IC50 and relative binding affinity (RBA) values. Overall, 188 chemicals were assayed, covering a million fold range of RBAs, from several different chemicals or used categories including steroidal estrogens, synthetic estrogens, antiestrogens, other miscellaneous steroids, alkylphenols, diphenyl derivatives, organochlorines, pesticides, alkyl hydroxybenzoate preservatives (parabens), phthalates, benzophenone compounds and a number of other miscellaneous chemicals. Of the 188 chemicals tested, 100 bound to the ER while 88 were non-binders. Included in the 100 chemicals that bound to the ER were 4-benzyloxyphenol, 2,4-dihydroxybenzophenone, and 2,2'-methylenebis(4-chlorophenol), compounds that have not been shown previously to bind the ER. It was also evident that certain structural features, such as an overall ring structure, were important for ER binding. These assays provided the most structurally diverse ER RBA data set with the widest range of RBA values published to date, and have met requirements for a designed training set from which to calibrate computational models to predict ER binding solely based on chemical structure.
The NCTR training set for ER binding comprises 130 active and 100 inactive compounds. To our best knowledge, this is the largest published estrogen dataset. This NCTR dataset has been extensively used to build and validate a series of computational models proposed for priority setting of potential estrogenic endocrine disruptors [Shi, 2001 (submitted) #225].
The androgen receptor (AR) assays used a validated (standardized) AR competitive binding assay to determine the affinity for a large, structurally diverse group of chemicals for the AR [Blair, 2000 #216]. The source of the AR for the competitive binding assay is a recombinant protein obtained from PanVera comprising the human AR ligand binding domain. The use of a pure AR binding domain eliminates any confounding assay conditions that might exist in assays using cytosolic tissue extracts that contain a myriad of soluble proteins.
Initially, test chemicals were screened at high concentrations to determine whether a chemical competed with [3H]-R1881 for the AR. Test chemicals that exhibited affinity for the AR in the first tier/screening were subsequently assayed using a wide range of concentrations to characterize the binding curve and to determine each chemical's IC50 and relative binding affinity (RBA) values. Overall, 204 chemicals were assayed, covering a 10 million fold range of RBAs, from several different chemical or use categories including natural and synthetic androgens, natural and synthetic estrogens, antiestrogens, other miscellaneous steroids, alkylphenols, diphenyl derivatives, organochlorines, pesticides, alkyl hydroxybenzoate preservatives (parabens), phthalates, benzophenone compounds and a number of other miscellaneous chemicals.
Of the 204 chemicals tested, 141 bound to the AR while 10 were slight binders
and 50 were non-binders. Comparison of these data to the binding affinity data
obtained from the estrogen receptor assays indicates that fewer chemicals bind
to the AR with high affinity. Conversely, fewer chemicals are non-binders in
the AR assay with a very high number of low affinity binders. These assays provided
the most structurally diverse AR RBA data set with the widest range of RBA values
published to date, and have met requirements for a designed training set from
which to calibrate computational models to predict AR binding solely based on
chemical structure.