Androgen receptor binding sites identified by a GREF_GATA model.

Abstract

Changes in transcriptional regulation can be permissive for tumor progression by allowing for selective growth advantage of tumor cells. Tumor suppressors can effectively inhibit this process. The PMEPA1 gene, a potent inhibitor of prostate cancer cell growth is an androgen-regulated gene. We addressed the question of whether or not androgen receptor can directly bind to specific PMEPA1 promoter upstream sequences. To test this hypothesis we extended in silico prediction of androgen receptor binding sites by a modeling approach and verified the actual binding by in vivo chromatin immunoprecipitation assay. Promoter upstream sequences of highly androgen-inducible genes were examined from microarray data of prostate cancer cells for transcription factor binding sites (TFBSs). Results were analyzed to formulate a model for the description of specific androgen receptor binding site context in these sequences. In silico analysis and subsequent experimental verification of the selected sequences suggested that a model that combined a GREF and a GATA TFBS was sufficient for predicting a class of functional androgen receptor binding sites. The GREF matrix family represents androgen receptor, glucocorticoid receptor and progesterone receptor binding sites and the GATA matrix family represents GATA binding protein 1-6 binding sites. We assessed the regulatory sequences of the PMEPA1 gene by comparing our model-based GREF_GATA predictions to weight matrix-based predictions. Androgen receptor binding to predicted promoter upstream sequences of the PMEPA1 gene was confirmed by chromatin immunoprecipitation assay. Our results suggested that androgen receptor binding to cognate elements was consistent with the GREF_GATA model. In contrast, using only single GREF weight matrices resulted in additional matches, apparently false positives. Our findings indicate that complex models based on datasets selected by biological function can be superior predictors as they recognize TFBSs in their functional context.

Full text

Page 1

COMMUNICATION
Androgen Receptor Binding Sites Identified by a
GREF_GATA Model
Katsuaki Masuda1, Thomas Werner2, Shilpi Maheshwari1
Matthias Frisch2, Soyon Oh1, Gyorgy Petrovics1, Klaus May2
Vasantha Srikantan1, Shiv Srivastava1and Albert Dobi1*
1Center for Prostate Disease
Research, Department of
Surgery, Uniformed Services
University, Rockville, MD
20852, USA
2Genomatix Software GmbH
D-80339 Munich, Germany
Changes in transcriptional regulation can be permissive for tumor
progression by allowing for selective growth advantage of tumor cells.
Tumor suppressors can effectively inhibit this process. The PMEPA1 gene, a
potent inhibitor of prostate cancer cell growth is an androgen-regulated
gene. We addressed the question of whether or not androgen receptor can
directly bind to specific PMEPA1 promoter upstream sequences. To test this
hypothesis we extended in silico prediction of androgen receptor binding
sites by a modeling approach and verified the actual binding by in vivo
chromatin immunoprecipitation assay. Promoter upstream sequences of
highly androgen-inducible genes were examined from microarray data of
prostate cancer cells for transcription factor binding sites (TFBSs). Results
were analyzed to formulate a model for the description of specific
androgen receptor binding site context in these sequences. In silico analysis
and subsequent experimental verification of the selected sequences
suggested that a model that combined a GREF and a GATA TFBS was
sufficient for predicting a class of functional androgen receptor binding
sites. The GREF matrix family represents androgen receptor, glucocorticoid
receptor and progesterone receptor binding sites and the GATA matrix
family represents GATA binding protein 1–6 binding sites. We assessed the
regulatory sequences of the PMEPA1 gene by comparing our model-based
GREF_GATA predictions to weight matrix-based predictions. Androgen
receptor binding to predicted promoter upstream sequences of the
PMEPA1 gene was confirmed by chromatin immunoprecipitation assay.
Our results suggested that androgen receptor binding to cognate elements
was consistent with the GREF_GATA model. In contrast, using only single
GREF weight matrices resulted in additional matches, apparently false
positives. Our findings indicate that complex models based on datasets
selected by biological function can be superior predictors as they recognize
TFBSs in their functional context.
q 2005 Elsevier Ltd. All rights reserved.
Keywords: model; androgen receptor; prostate cancer; promoter database;
GATA factor
*Corresponding author
Alterations in transcription regulation are hall-
marks of cancer progression. Decreases in the
expression of specific genes that control cell
proliferation may significantly contribute to malig-
nant transformation. The PMEPA1 gene has been
recognized as a potent cell growth inhibitor of
prostate cancer cells,1with decreased expression of
PMEPA1 associated with higher pathologic stage in
prostate cancer specimens. Function for PMEPA1 in
the ubiquitin-proteasome pathway is suggested by
shared structural similarities between PMEPA1 and
0022-2836/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
Abbreviations used: TFBS, transcription factor binding
site; AR, androgen receptor; ChIP, chromatin immuno-
precipitation; GREF, glucocorticoid responsive element
matrix family; ARE, androgen responsive element.
E-mail address of the corresponding author:
adobi@cpdr.org
doi:10.1016/j.jmb.2005.09.009J. Mol. Biol. (2005) 353, 763–771

Page 2

Nedd4-binding proteins. Interestingly, PMEPA1
gene expression was shown to be regulated by
androgens in prostate cancer cells.2–4In order to
assessandrogenreceptor
regulatory elements of the PMEPA1 gene we
integrated computation biology with wetlab experi-
mentation. Computation molecular biology of
gene regulation is an emerging frontier of bio-
informatics.5In the past, conclusive identification of
functionally active transcription factor binding sites
(TFBSs) relied mainly on laborious wetlab experi-
mentations. However, recently chromatin immuno-
precipitation (ChIP)-based assays such as “ChIP-
chip” and ChIP Display were shown to be robust
techniques to assess binding sites experimentally
genome-wide.6,7In support to these efforts, robust
platforms are currently available for in vitro
selection of single TFBSs to assess optimal cognate
sites for various elements.8–10Although bio-
informatic tools are available for the prediction of
TFBSs, we preferred a weight matrix approach.11
However, the application of any weight matrix
approach to long sequences is hampered by a large
number of false positive predictions. Although
examining the evolutional conservation of tran-
scription factor binding can enhance the specificity
of in silico predictions,12,13this technique can be
confounded by species-specific genes. Notably,
many human cancer-related genes as illustrated
by prostate specific antigen (PSA, KLK3) barely
resemble their orthologs from other mammalian
species.14Among mammals the pathogenesis of
prostate cancer (CaP) is unique to man. The only
mammal known to develop CaP in a similar fashion
is the canine.15Fortunately, biological context
can significantly enhance the quality of TFBS
models.16,17Biologically linked sets of transcription
factors observed in the combination of specific
TFBSs can significantly improve prediction quality.5
Consistent with this notion, we selected a training
set of genes based on gene expression kinetics and
dose–response of prostate cancer cells to androgen
hormone stimulus.4Furthermore, we reasoned that
gene regulation by the combination of closely
situated binding sites is an essential feature of
functional transcription regulatory regions.18,19
Therefore, we formulated an androgen receptor-
binding model from the combination of adjacent
matrices. We experimentally verified this model
and went on to assess the utility of the model by
testing upstream sequences of the androgen-
inducible gene, PMEPA1.2
(AR)bindingto
Defining a model for androgen receptor binding
in the context of prostate-specific hormone
induction
Upstream
inducible genes were selected as a training set
from our previous oligonucleotide array experi-
ments in orderto formulate a model that can predict
functional androgen receptor binding sites in the
context of androgen-induced prostate cancer cells
(Table 1). Four genes were selected that stood out of
the list of androgen-inducible genes both by the
robustness of gene expression response (KLK3Z
58! and NDRG1Z13!) and by kinetic consider-
ations (IHPK1Z9! in 1 h and NKX3.1Z5! in
6 h).4
Extraction of 50upstream sequences relative to
the transcription start sites of training set genes
between positions K2000 and C1 was accom-
plished by the ElDorado system (Genomatix,
Munich)† that is based on the genome assemblies
from NCBI (Homo sapiens NCBI Build 35‡).
Sequences were verified by database searches in
the human genome. In this process upstream
sequences of KLK3 and KLK2 genes were found to
show high degrees of identity (approx. 80% within
K1000;C1). Although, we previously found that
KLK2 was a highly androgen-inducible gene, KLK2
was excluded from the training set to avoid
sequence redundancy that may introduce bias into
the model prediction. The training set was analyzed
by the MatInspector20to identify potential andro-
gen receptor binding sites. Androgen receptor can
recognize DNA sequences represented by the
glucocorticoid responsive element matrix family
(GREF) that includes matrices of androgen receptor,
glucocorticoid receptor and progesterone receptor
binding sites. Two GREFs in KLK3 were identified,
one in NDRG1, two in NKX3.1 and one within the
IHPK1 upstream sequences. In addition to the
GREF family the program found 27 other matrix
families to match within the sequences. The relation
of each predicted site was assessed relative to
the GREF positions. The distance criterion for the
assessment was 20–33 base-pairs between the
centers of two adjacent sites (within 2–3 DNA
helical turn distance). Only the GATA matrix family
matches showed distance correlation to the GREF
sequencesofhighlyandrogen-
Table 1. Androgen-inducible gene training set
GeneAccessionLocusContigAbsolute positions (strand)
KLK3 (PSA)
NDRG1
NKX3.1
IHPK1
HT2351
D87953
U80669
D87452
19q13.41
8q24.3
8p21
3p21.31
NM_145864
NT_008046
NT_023666
NT_022517
23,624,361–23,626,361 (C)
47,529,674–47,527,674 (K)
1,916,763–1,914,763 (K)
49,711,601–49,709,601 (K)
Gene symbols, accession numbers and chromosomal locations (Homo sapiens NCBI Build 35, http://www.ncbi.nlm.nih.gov) are shown.
† http://www.genomatix.de
‡ http://www.ncbi.nlm.nih.gov
764
Predicting Androgen Receptor Binding Sites

Page 3

matrix family matches (Figure 1). Indeed, the
identification of potential GATA binding sites
close to androgen responsive elements is consistent
with previous findings that suggested critical roles
for GATA2 and GATA3 transcription factors in
androgen receptor functions.21It has been demon-
strated that GATA-driven expression of SV40
T-antigen resulted in androgen-independent pros-
tate cancer in transgenic mice.22Moreover, the KLK3
gene upstream androgen responsive enhancer
(AREIII) (Figure 2) that is embedded in a cluster
of adjacent GATA factor binding sites was shown to
be essential for optimal androgen induction of KLK3
gene expression.22
Experimental validation of the GREF_GATA
model
By assessing various core similarities we found
that a threshold of 0.85 core similarity was optimal
for both matrices. Matrix similarities were set to
default values (optimized). The model was then
defined using the FastM module of GEMS
Launcher13and was successfully checked for
correctness by searching the sequences of the
training data set by ModelInspector.23To compen-
sate for model overfitting we assessed the validity
of our findings by an independent experimental
approach. For that purpose we tested the predicted
sites in our training set for androgen receptor
binding in androgen-induced prostate cancer cells
by in vivo ChIP. The KLK3 gene androgen respon-
sive enhancer (ARE III) was assayed as an
additional positive control (Figure 2). Hormone-
induced androgen receptor binding to predicted
sites was evident in four out of the five tested gene
promoter upstream sequences.
Androgen receptor binds preferentially to the
region defined by the GREF_GATA model within
the PMEPA1 gene promoter upstream
sequences in androgen-induced LNCaP cells
We used the experimentally supported GREF_
GATA model to analyze the upstream K8000;C1
sequencesof theandrogen-inducible gene,
Figure 1. Matrices within the
training set sequences. (a) Position
of matches of GREF (encircled
filled arrows) and GATA (open
arrowheads) matrix
withinthetraining
sequences. Arrowheads indicate
the strand orientation. Adjacent
GREF and GATA matrices (min.
20and max.
between the centers of matrices)
and the corresponding matrix
anchorpositions
(b) List of all matches of matrix
families identified within the train-
ing set sequences. Numbers indi-
cate how often a given matrix
matched within the pre-defined
range of a GREF matrix match
(20–33 bp). Only GATA matrix
family matches fulfilled this con-
dition in all four sequences.
families
set of
33 bpinterval
are boxed.
Predicting Androgen Receptor Binding Sites
765

Page 4

PMEPA1. The 50distal boundary of the selected
region (K8000) excluded a cluster of Alu repeats
and a matrix attachment region (S/MAR) and the
proximal boundary was extended until the tran-
scription initiation site (C1). Two model matches at
positions K2134 (GREF)/K2156 (GATA) and K230
(GREF)/K250 (GATA) were found with model
scores
(Figure 3(a)). In contrast, eight GREF matrix family
matches were found that included preferred sites of
androgen,progesterone
responsive elements (ARE, PRE, and GRE,
respectively). In vivo binding of androgen receptor
to the PMEPA1 gene upstream sequences in LNCaP
of 95.1% and92.6%,respectively
andglucocorticoid-
Figure 2. Experimental verifica-
tion ofpredicted
receptor binding sites. (a) ChIP
assay results of androgen-induci-
ble gene promoters are shown.
GREF_GATA model-match pos-
itions are indicated under gene
symbols. For ChIP assay, prostate
cancer cells (LNCaP, American
Type Culture Collection, Rockville,
MD) were cultured in charcoal
treated androgen-free fetal bovine
serum (Invitrogen, Carlsbad, CA)
for five days to deplete androgens
from the cells. After androgen
depletion cell cultures were incu-
bated in the presence or absence of
3.3 nM of R1881 non-metabolizable
androgen (DuPont, Boston, MA)
for 24 h. For ChIP assay the
medium was removed from the
cells by aspiration. In each parallel,
2!106LNCaP cells were incubated
in PBS buffer with 1% formal-
androgen
dehyde (Sigma, St. Louis, MO) at 0 8C for 10 min followed by incubation at 37 8C for 15 min. The crosslinking reaction
was stopped by the addition of glycine to a final concentration of 0.125 M and the cells were washed three times with ice-
cold 1xPBS in the presence of Complete Mini protease inhibitor (Roche, Indianapolis, IN) and Phosphatase Inhibitor
Cocktail I and II (Sigma, St. Louis, MO). Inhibitors were added to the reaction mixtures in all subsequent steps. The cells
were lysed (50 mM Tris–HCl (pH 8.0), 1% (w/v) SDS, 10 mM EDTA) and the lysates were incubated at 0 8C for 10 min.
The chromatin was fragmented by sonicating the lysates with a VirSonic 100 sonicator microtip (VIRTIS, Gardiner, NY)
applying four times 10 s bursts at energy level 4, followed by 50 s of cooling after each burst. The cell debris were
separated by centrifugation at 10,000g at 4 8C in a microcentrifuge and the supernatants were diluted ten times (16.7 mM
Tris–HCl (pH 8.0), 167 mM NaCl, 1.1% TritonX-100, 1.2 mM Na-EDTA, 0.01% SDS). The suspensions were filtrated
through 0.45 mm filter syringes and the flow-through was collected in pre-chilled tubes. Aliquots of 50 ml of 50%
immobilized and pre-blocked protein A/G-bead slurry (PIERCE, Rockford, IL) were added to an aliquot of 250 ml of the
filtrates and the sample tubes were agitated on a head-to-head rotator for 1 h at 4 8C. Subsequently the beads were
removed by centrifugation (4 8C, 3000g, 2 min) and 200 ml of the clear supernatants were transferred into pre-chilled
microcentrifuge tubes. To each tube, 2 ml (0.33 mg/ml) of anti-androgen receptor antibody AR(PG21) (Upstate, Lake
Placid, NY) was added. The immunoprecipitation reactions were incubated for 12 h at 4 8C with gentle head-to-head
rotation. Subsequently, the chromatin–antibody complexes were captured by the addition of 60 ml of immobilized
protein A/G slurry by incubating the suspensions at 4 8C (1 h) continuing gentle head-to-head rotation. After incubation
the beads were collected by brief centrifugation and the beads were washed four times with ice-cold buffer of 20 mM
Tris–HCl (pH 8.0), 150 mM NaCl, 2 mM Na-EDTA, 0.1% SDS and 1% TritonX-100, once at 24 8C with the buffer of 20 mM
Tris–HCl (pH 8.0), 250 mM LiCl, 1 mM Na-EDTA, 0.5% IGEPAL (Sigma, St. Louis, MO) non-ionic detergent and 0.5%
sodium-deoxycholate (Sigma, St. Louis, MO) and twice with the solution of 10 mM Tris–HCl (pH 8.0), 1 mM Na-EDTA at
room temperature. After each washing step the beads were collected by centrifugation (24 8C, 1 min, 4000g). The beads
were resuspended in TE-buffer (10 mM Tris–HCl (pH 7.5), 1 mM Na-EDTA) containing 0.5% of SDS and Proteinase K
enzyme (Roche, Indianapolis, IN) to a final concentration of 0.5 mg/ml to digest the protein components. The reaction
mixtures were incubated for 12 h at 37 8C. The clear supernatants were recovered by spinning the sample in a
microcentrifuge and the suspensions were incubated at 65 8C for 6 h to revert the chemical crosslinking. Then the
samples were purified by phenol/chloroform extraction and subsequent precipitation with ethanol. The pellets were
dissolved in TE buffer. The input genomic DNA samples were treated in the same way except that no
immunoprecipitation was performed. (b) Specific sets of primer pairs were selected for PCR reactions in order to
amplify the regions of predicted GREF sequences within the promoter upstream sequences of KLK3, NDRG1, NKX3.1
and IHPK1 genes. The amplification was carried out in a T-Gradient Thermoblock (Biometra, Tampa, FL) (94 8C, 15 s;
annealing at 58 8C (for KLK3/AREIII 55 8C); 30 s, 72 8C, 1 min). To achieve linear amplification ranges ChIP and input
products were amplified with 43 and 38 cycles, respectively.
766
Predicting Androgen Receptor Binding Sites

Page 5

prostate cancer cells in response to R1881 synthetic
androgen stimulus was examined. Androgen
receptor binding was tested by ChIP assay using
anti-AR antibody and primer sets to amplify the
predicted binding sites of GREF matrix matches
within the K8000;C1 region of the PMEPA1 gene
(Figure 3(b)). Our GREF_GATA model correctly
predicted binding of androgen receptor to its
cognate elements in response to androgen induc-
tion (Figure 3(c)). It was surprising that a seemingly
excellent GREF/ARE matrix match (K7168 in
Figure 3) was not a functional androgen receptor
binding site as opposed to the two predicted
GREF_GATA models, both of which could be
validated by ChIP. It is remarkable that both sites
were recognized as glucocorticoid and progester-
one receptor binding sites, demonstrating that the
matrix family concept is valid (as shown by the
chromatin immunoprecipitation) as well as essen-
tial for model definition.
Promoter database search with the GREF_GATA
model
The model has shown good predictive power in
the case of the PMEPA1 gene. Therefore, the entire
Figure 3. Positions of model and matrix-matches and experimental validation of corresponding androgen receptor
binding sites within the PMEPA1 gene upstream sequences. (a) GREF_GATA model-matches are shown in bold.
Predicted matches for GREF matrix family members ARE, GRE, and PRE, respectively, are shown below. (b) Sequences
of forward (F) and reverse (R) primer pairs and corresponding annealing temperatures are listed. (c) ChIP results with
anti-androgen receptor antibodies and the control input amplification products (Input) are shown. Prostate cancer cells
(LNCaP) were incubated in the absence (K) or presence (C) of synthetic androgen (R1881). For details see the legend to
Figure 2.
Predicting Androgen Receptor Binding Sites
767

Page 6

Genomatix Promoter Database (GPD) was searched
for promoters from human annotated genes with
the GREF_GATA model (ModelInspector, default
settings). The search identified 995 matches
(belonging to 933 different genes) out of 36,943
promoters analyzed, a match rate of 2.7%, indicat-
ing moderate selectivity. Statistical analysis of the
933 genes revealed that the selection of identified
genes was not random and the gene ontology (GO)
category “sensory perception of chemical stimulus”
was clearly over-represented in the match list. This
most significant group in the ontology “biological
process” (z-score 6.19) contained nine genes, seven
of which were olfactory receptors. Within the
ontology “molecular function” the group “olfactory
receptor activity” was the top-scoring group
(z-score 7.09) containing more than five genes.
Further inspection of the output revealed 18 genes
out of a total of 270 olfactory receptor genes in the
human genome (numbers taken from ElDorado).
Thus olfactory receptors were 2.5-fold more often
found than expected at random. Promoters of nine
genes of the top-scoring group were aligned using
the program DiAlign to exclude the possibility that
observed matches were due to extensive sequence
similarities due to recent gene/promoter dupli-
cation.24The resulting sequence similarities were
(except for the pair OR5I1/OR6C3, 32%) all below
the random similarity of 25%, strongly suggesting
that the conservation of the promoter framework
was due to functional constraints rather than
sequence similarity. Assessment of specificity and
sensitivity would require knowledge of the absolute
number of true positive matches. This obstacle was
circumvented by assessing whether the match list
was enriched in potentially androgen receptor
associated genes or not, an approach feasible
based on available data. The total number of
genes co-cited with androgen receptor in the
literature is 1034. Co-citation analysis was done
with BiblioSpere†, a literature-mining tool based
on PubMed abstracts‡. A total of 51 of the 933
genes identified by the GREF_GATA model were
co-cited with androgen receptor. To determine if
these results indicate a significant association
with androgen receptor, 30 random picks of
genes were identified in the database. The
number ofgenesco-cited
receptor in the literature within the random
group was determined. A z-score (number of
genes co-cited with AR in model search–average
number of genes co-cited with AR in random
picks) / standard deviation of random pick
results was calculated. The resulting z-score
was 4.4, indicating a significant enrichment of
androgen receptor co-cited genes in the model-
search result.
For model validation we assessed the binding
of androgen receptor to the promoter sequences
withandrogen
of genes within the prostate cancer biological
context by in vivo ChIP assay. The assay
indicated androgen receptor binding to six out
of eight tested gene promoter elements in
response toandrogen
LNCaP cells (Figure 4). These data suggested
that the GREF_GATA model was a powerful
predictorofbiologically
receptor binding sites.
(R1881)stimulusin
activeandrogen
Biological context is an important factor in
training set selection
Androgens and their receptor are the main
therapeutic targets in prostate cancer.25While
androgen blockade has been a significant thera-
peutic choice,26failure of androgen-blockade
therapy can dramatically reduce the survival
rate.27Identification of specific AR cognate binding
sequences throughout the entire human genome
relying exclusively on experimental high through-
put TFBS identification techniques may not be
practical. A high rate of missed TFBS in robust
scanning techniques has been reported.28On the
other hand, bioinformatic tools are readily available
to predict potential androgen receptor binding sites.
However,all single
approaches areplagued by an enormous proportion
of presumably false positive predictions, referred to
as the “futility theorem”.5Therefore, we addressed
the question of whether or not androgen receptor
binding sites can be more selectively predicted by a
model-based approach. We found that a model
composed of glucocorticoid responsive element
matrices (GREF) adjacent to GATA matrices
(GATA) allowed the identification of a novel class
of biologically active androgen receptor binding
sites. Selection of a training set of promoters from a
defined biological context was crucial for successful
model generation.29,30A genome-wide promoter
search with the GREF_GATA model has identified
933 genes, 51 of which were found to be co-cited
with androgen receptor in the literature. Five
examples of known androgen/prostate cancer
related genes are: six transmemebrane epithelial
antigen of prostate 2, STEAP2 (Loc261729);31acid
phosphatase of prostate, ACPP (Loc55);32breast
cancer 2 early onset, BRCA2 (Loc 675);33aconitase 2
mitocondrial, ACO2 (Loc50);34androgen-induced
proliferation inhibitor, APRIN (Loc23047).35It is
interesting to note that the top scoring GO-category
of olfactory receptors was functionally linked to
androgen receptor in a previous study.36These
findings are consistent with the known role that
androgen receptor plays in gender-specific sensory
functions.37–39
In summary, we selected genes based on their
expressionresponseto
induction before defining our training set. The
hormone-responsive prostate cancer cell line
LNCaP as experimental host was selected for its
wide use as an experimental model.40Our previous
oligonucleotide array expression platform directly
bindingsiteoriented
androgenhormone
† http://www.genomatix.de
‡ http://www.ncbi.nlm.nih.gov/Entrez
768
Predicting Androgen Receptor Binding Sites

Page 7

revealed the time kinetic- and dose-dependent
responses of gene expression profiles. The KLK3,
NDRG1, NKX3.1, and IHPK1 genes were selected as
our training set based on this biological evidence.
We have defined GREF_GATA, a novel model to
predict androgen responsive elements that will
provide new insights into the androgen-regulated
transcriptome in prostate cancer. Identifying direct
targets of AR will allow the detection of critical
changes of androgen signaling in prostate cancer
progression.
Acknowledgements
We are thankful for the support of Dr David G.
McLeod, Director of the Center for Prostate Disease
Research. Also, we thank the suggestions of Drs
Linda Xu and Eric Richter. This research was
supported by the Center for Prostate Disease
Research Program through the Henry M. Jackson
Foundation for the Advancement of Military
Medicine under contract number HU001-04-C-
1502 (2004) with the Uniformed Services University.
Materials in this research were all commercially
available products. The opinions and assertions
contained herein are the private views of the
authors and are not to be considered as reflecting
the views of the Henry M. Jackson Foundation for
the Advancement of Military Medicine or the US
Department of Defense.
Supplementary Data
Supplementary
article can be found, in the online version, at
doi:10.1016/j.jmb.2005.09.009
data associatedwith this
References
1. Xu, L. L., Shi, Y., Petrovics, G., Sun, C., Makarem, M.,
Zhang, W. et al. (2003). PMEPA1, an androgen-
regulated NEDD4-binding
growth inhibitory function and decreased expression
during prostate cancer progression. Cancer Res. 63,
4299–4304.
2. Xu, L. L., Shanmugam, N., Segawa, T., Sesterhenn,
I. A., McLeod, D. G., Moul, J. W. et al. (2000). A novel
androgen-regulatedgene,
chromosome 20q13 exhibits high level expression in
prostate. Genomics, 66, 257–263.
3. Xu, L. L., Su, Y. P., Labiche, R., Segawa, T.,
Shanmugam, N., McLeod, D. G. et al. (2001).
Quantitative expression
regulated genes in prostate cancer cells and identifi-
cation of prostate-specific genes. Int. J. Cancer, 92,
322–328.
4. Segawa, T., Nau, M. E., Xu, L. L., Chilukuri, R. N.,
Makarem, M., Zhang, W. et al. (2002). Androgen-
protein,exhibits cell
PMEPA1,locatedon
profileof androgen-
Figure 4. Hormone-induced
androgen receptor binding to pre-
dicted binding sites in the genome
of prostate cancer cells. (a) ChIP
assay of STEAP2, ACPP BRCA2,
ACO2, APRIN, ACLY, SOX5 and
SOX6 gene promoters. GREF_-
GATA model-match positions are
shownunder
Model-match was found down-
stream relative to an alternative
transcription initiation site of the
APRIN gene. LNCaP cells were
incubated in the presence (C) or
absence (K) of R1881 synthetic
androgen. Input
amplification product of control
genomic DNA. In vivo ChIP assay
was carried out as described in the
legend to Figure 2. (b) Primer
sequences and positions relative
to the transcription initiation sites
of the tested genesarelisted. Inthe
amplification step all primers
were annealed at 58 8C.
genesymbols.
marks the
Predicting Androgen Receptor Binding Sites
769

Page 8

induced expression of endoplasmic reticulum (ER)
stress response genes in prostate cancer cells.
Oncogene, 21, 8749–8758.
5. Wasserman, W. W. & Sandelin, A. (2004). Applied
bioinformatics for the identification of regulatory
elements. Nature Rev. Genet. 5, 276–287.
6. Iyer, V. R., Horak, C. E., Scafe, C. S., Botstein, D.,
Snyder, M. & Brown, P. O. (2001). Genomic binding
sites of the yeast cell-cycle transcription factors SBF
and MBF. Nature, 409, 533–538.
7. Barski, A. & Frenkel, B. (2004). ChIP Display: novel
method for identification of genomic targets of
transcription factors. Nucl. Acids Res. 32, e104.
8. Bulyk, M. L., Gentalen, E., Lockhart, D. J. & Church,
G. M. (1999). Quantifying DNA-protein interactions
by double-stranded DNA arrays. Nature Biotechnol.
17, 573–577.
9. Roulet, E., Busso, S., Camargo, A. A., Simpson, A. J.,
Mermod, N. & Bucher, P. (2002). High-throughput
SELEX SAGE method for quantitative modeling of
transcription-factor binding sites. Nature Biotechnol.
20, 831–835.
10. Linnell, J., Mott, R., Field, S., Kwiatkowski, D. P.,
Ragoussis, J. & Udalova, I. A. (2004). Quantitative
high-throughput analysis of transcription factor
binding specificities. Nucl. Acids Res. 32, e44.
11. Cartharius, K., Frech, K., Grote, K., Klocke, B.,
Haltmeier, M., Klingenhoff,
MatInspector and beyond: promoter analysis based
on transcription factor binding sites. Bioinformatics, 21,
2933–2942.
12. Dieterich, C., Rahmann, S. & Vingron, M. (2004).
Functional inference from non-random distributions
of conserved predicted transcription factor binding
sites. Bioinformatics, 20, I109–I115.
13. Klingenhoff, A., Frech, K., Quandt, K. & Werner, T.
(1999).Functional promoter
detected by formal models independent of overall
nucleotide sequence similarity. Bioinformatics, 15,
180–186.
14. Olsson, A. Y., Lilja, H. & Lundwall, A. (2004). Taxon-
specific evolution of glandular kallikrein genes and
identification of a progenitor of prostate-specific
antigen. Genomics, 84, 147–156.
15. Russell, P. J. & Voeks, D. J. (2003). Animal models of
prostate cancer. Methods Mol. Med. 81, 89–112.
16. Fessele, S., Maier, H., Zischek, C., Nelson, P. J. &
Werner, T. (2002). Regulatory context is a crucial part
of gene function. Trends Genet. 18, 60–63.
17. Werner, T., Fessele, S., Maier, H. & Nelson, P. J. (2003).
Computer modeling of promoter organization as a
tool to study transcriptional coregulation. FASEB J. 17,
1228–1237.
18. Kel-Margoulis,O.V.,
Kolchanov, N. A., Wingender, E. & Kel, A. E. (2000).
COMPEL: a database on composite regulatory
elements providing combinatorial transcriptional
regulation. Nucl. Acids Res. 28, 311–315.
19. Werner, T. (2001). The promoter connection. Nature
Genet. 29, 105–106.
20. Quandt, K., Frech, K., Karas, H., Wingender, E. &
Werner, T. (1995). MatInd and MatInspector: new
fast and versatile tools for detection of consensus
matches in nucleotide sequence data. Nucl. Acids Res.
23, 4878–4884.
21. van den Bemd, G. J., Jhamai, M., Brinkmann, A. O. &
Chang, G. T. (2003). The atypical GATA proteinTRPS1
A.etal. (2005).
modulescanbe
Romashchenko,A. G.,
represses androgen-induced prostate-specific antigen
expression in LNCaP prostate cancer cells. Biochem.
Biophys. Res. Commun. 312, 578–584.
22. Perez-Stable, C. M., Pozas, A. & Roos, B. A. (2000).
A role for GATA transcription factors in the androgen
regulation of the prostate-specific antigen gene
enhancer. Mol. Cell. Endocrinol. 167, 43–53.
23. Morgenstern, B., Dress, A. & Werner, T. (1996).
Multiple DNA and protein sequence alignment
based on segment-to-segment comparison. Proc. Natl
Acad. Sci. USA, 93, 12098–12103.
24. Morgenstern, B. (2004). DIALIGN: multiple DNA and
protein sequence alignment at BiBiServ. Nucl. Acids
Res. 32, W33–W36.
25. Chen, C. D., Welsbie, D. S., Tran, C., Baek, S. H., Chen,
R., Vessella, R. et al. (2004). Molecular determinants of
resistance to antiandrogen therapy. Nature Med. 10,
33–39.
26. Moul, J. W. (1998). Contemporary hormonal manage-
mentof advancedprostate
(Huntingt), 12, 499–508.
27. Huang, H. & Tindall, D. J. (2002). The role of the
androgen receptor in prostate cancer. Crit. Rev.
Eukaryot. Gene Expr. 12, 193–207.
28. Buck, M. J. & Lieb, J. D. (2004). ChIP-chip: consider-
ations for the design, analysis, and application of
genome-widechromatin
experiments. Genomics, 83, 349–360.
29. Krivan, W. & Wasserman, W. W. (2001). A predictive
model for regulatory sequences directing liver-
specific transcription. Genome Res. 11, 1559–1566.
30. Liu, R., McEachin, R. C. & States, D. J. (2003).
Computationally identifying novel NF-kappa B-regu-
lated immune genes in the human genome. Genome
Res. 13, 654–661.
31. Porkka, K. P., Helenius, M. A. & Visakorpi, T. (2002).
Cloning and characterization of a novel six-trans-
membrane protein STEAP2, expressed in normal and
malignant prostate. Lab. Invest. 82, 1573–1582.
32. Winqvist, R., Virkkunen, P., Grzeschik, K. H. & Vihko,
P. (1989). Chromosomal localization to 3q21–qter and
two TaqI RFLPs of the human prostate-specific acid
phosphatase gene (ACPP). Cytogenet. Cell Genet. 52,
68–71.
33. Simard, J., Dumont, M., Labuda, D., Sinnett, D.,
Meloche, C., El-Alfy, M. et al. (2003). Prostate cancer
susceptibility genes: lessons learned and challenges
posed. Endocr. Relat. Cancer, 10, 225–259.
34. Juang, H. H. (2004). Nitroprusside stimulates mito-
chondrial aconitase gene expression through the
cyclic adenosine 30,50-monosphosphate signal trans-
duction pathway in human prostate carcinoma cells.
Prostate, 61, 92–102.
35. Geck, P., Szelei, J., Jimenez, J., Sonnenschein, C. &
Soto, A. M. (1999). Early gene expression during
androgen-inducedinhibition
prostate cancer cells: a new suppressor candidate on
chromosome 13, in the BRCA2-Rb1 locus. J. Steroid.
Biochem. Mol. Biol. 68, 41–50.
36. Paredes, R. G., Lopez, M. E. & Baum, M. J. (1998).
Testosterone augments neuronal Fos responses to
estrous odors throughout the vomeronasal projection
pathway of gonadectomized male and female rats.
Horm. Behav. 33, 48–57.
37. Cooke, B. M., Breedlove, S. M. & Jordan, C. L. (2003).
Both estrogen receptors and androgen receptors
contribute to testosterone-induced changes in the
morphology of the medial amygdala and sexual
arousal in male rats. Horm. Behav. 43, 336–346.
cancer.Oncology
immunoprecipitation
ofproliferation of
770
Predicting Androgen Receptor Binding Sites

Page 9

38. Zakon, H. (2004). Physiology. Heeding the hormonal
call. Science, 305, 349–350.
39. Kritzer, M. (2004). The distribution of immuno-
reactivity for intracellular androgen receptors in the
cerebral cortex of hormonally intact adult male and
femalerats:localization inpyramidalneurons
making corticocortical connections. Cereb. Cortex, 14,
268–280.
40. Horoszewicz, J. S., Leong, S. S., Chu, T. M., Wajsman,
Z. L., Friedman, M., Papsidero, L. et al. (1980). The
LNCaP cell line—a new model for studies on human
prostatic carcinoma. Prog. Clin. Biol. Res. 37, 115–132.
Edited by M. Yaniv
(Received 11 April 2005; received in revised form 31 July 2005; accepted 7 September 2005)
Available online 22 September 2005
Predicting Androgen Receptor Binding Sites
771