The Response of the Prostate to Circulating Cholesterol: Activating Transcription Factor 3 (ATF3) as a Prominent Node in a Cholesterol-Sensing Network

Elevated circulating cholesterol is a systemic risk factor for cardiovascular disease and metabolic syndrome, however the manner in which the normal prostate responds to variations in cholesterol levels is poorly understood. In this study we addressed the molecular and cellular effects of elevated and suppressed levels of circulating cholesterol on the normal prostate. Integrated bioinformatic analysis was performed using DNA microarray data from two experimental formats: (1) ventral prostate from male mice with chronically elevated circulating cholesterol and (2) human prostate cells exposed acutely to cholesterol depletion. A cholesterol-sensitive gene expression network was constructed from these data and the transcription factor ATF3 was identified as a prominent node in the network. Validation experiments confirmed that elevated cholesterol reduced ATF3 expression and enhanced proliferation of prostate cells, while cholesterol depletion increased ATF3 levels and inhibited proliferation. Cholesterol reduction in vivo alleviated dense lymphomononuclear infiltrates in the periprostatic adipose tissue, which were closely associated with nerve tracts and blood vessels. These findings open new perspectives on the role of cholesterol in prostate health, and provide a novel role for ATF3, and associated proteins within a large signaling network, as a cholesterol-sensing mechanism.


Introduction
The human prostate is subject to a variety of pathologic conditions and syndromes that are not well understood in molecular terms. Most investigators have concluded that prostate health, particularly with respect to prostate cancer, is susceptible to lifestyle influences [1]. The association with lifestyle likely reflects a complex interplay between genetic, epigenetic, biochemical and metabolic processes. This is particularly evident in the US where sedentary habits and an obesogenic diet are now widespread. Accumulating evidence indicates that heart disease, diabetes and metabolic syndrome are associated with increased risk or severity of lower urinary tract symptoms (LUTS) [2][3][4].
Cholesterol, and structurally related lipids, are critical in membrane assembly and integrity. All eukaryotic cell membranes have cholesterol, ergosterol, or a phytosterol as a major membrane component. Cholesterol is one of the key regulators of membrane dynamics, a role tied to its tendency to facilitate the close packing of saturated acyl chains of membrane phospholipids, thereby stabilizing local membrane structure. Consequently, membrane compartmentalization into functional subdomains influenced by cholesterol concentration provides for post-translational control over important signal transduction pathways. The prostate synthesizes high levels of cholesterol, at similar rates to the liver, and the prostate accumulates cholesterol deposits with age [5]. Recent evidence from pre-clinical models has demonstrated a role for cholesterol in signal transduction in prostate cancer cells and tissues [6,7], as well as in intraprostatic/ intratumoral steroidogenesis [8] consistent with epidemiologic studies indicating that high circulating cholesterol promotes aggressive forms of the disease [9][10][11][12].
Normal tissues sense and respond to variations in circulating cholesterol. The adult prostate gradually loses the ability to regulate cholesterol levels at normal homeostatic levels, resulting in accumulation of excess intraprostatic cholesterol [5]. The degree or manner of the effect of this accumulation on prostate physiology is poorly understood. One community-based cohort study found a 4-fold increased risk of benign prostatic hyperplasia (BPH) among diabetic men with low density lipoprotein (LDL) cholesterol in the highest tertile in comparison to men in the lowest tertile [13]. A recent study from our group demonstrated that age-related prostate enlargement in the Syrian 87.20 hamster was dependent on the presence of cholesterol in the diet, and could be reversed with ezetimibe, a specific inhibitor of cholesterol absorption from the intestine [14]. However, our understanding of the effects of elevated cholesterol and hypocholesterolemic drugs on the prostate is limited. It is not known, for example, whether hypercholesterolemia elicits physiologic, metabolic, or biochemical changes in the prostate that predispose toward disease.
In the present study we attempted to uncover a cholesterolsensitive signaling network in the normal prostate of the mouse and human prostate cells using a systems approach. Our findings indicate that the prostate responds to variations in circulating cholesterol levels by altering cholesterol tissue content, cell proliferation rate and gene expression. Evidence also suggests that hypocholesterolemia may suppress prostatic inflammation. Our findings provide the first broad look at the manner in which the normal prostate responds to changes in circulating cholesterol.

Animal Maintenance and Diet
Male SCID and C57BL/6 mice (5-6 wk old) were obtained from the Massachusetts General Hospital and Jackson Labs, respectively. All mice were maintained on a low fat, no cholesterol diet (LFNC) (Research Diets, New Brunswick, NJ diet # D12102), equivalent to normal mouse chow, but without lot to lot variation for 2 weeks to normalize cholesterol levels, after which mice were divided into individual diet groups. C57BL/6 mice were divided into 3 diet groups (n = 18/group): hypocholesterolemic (Hypo) (LFNC + ezetimibe-30mg/kg/day; Schering-Plough, New Brunswick, NJ) added to the food), normocholesterolemic (Normo) (LFNC diet) or hypercholesterolemic (Hyper) (high fat, high cholesterol (HFHC) diet) (Research Diets, diet # D12108). SCID mice were grouped into 2 diet groups; Hyper and Normo. Animals were maintained on the respective diets for 4 months. Animals were housed in compliance with Children's Hospital Boston's animal care and use policies. All procedures were approved by IACUC under protocols A07-06-084 and 07-08-1503. Terminal bleeds were taken for liver function testing and serology by cardiac puncture. Prostates were removed and divided into individual anatomic lobes, i.e. ventral (VP), dorsal lateral (DLP) and anterior (AP) under a dissecting microscope, and either placed in OCT solution (Tissue-Tek, Torrance, CA) or in formalin for paraffin embedding [15].
Immunohistochemistry and Immunofluorescence Cell Staining 4-mm cryostat-obtained tissue sections were mounted on Superfrost/Plus microscope slides and kept at -80uC before use. Frozen sections were kept at RT for 1 min, followed by 3X washes in PBS. Fixation was done with Permeabilization Solution (BD Bioscience, San Diego, CA) for 15 min. After 1 h blocking with 3% BSA/PBS solution, slides were incubated for 12 h at 4uC with specific antibodies, followed by incubation with HRP-conjugated secondary IgG for 30 min at RT. After washing with PBS (3X), 3-3' diaminobenzidine (DAB) substrate chromogen solution (Envision Plus Kit, Dako Corp) was applied. The reaction was monitored by microscopy and was terminated when properly developed. Nuclei were counterstained with Meyer's hematoxylin and slides were analyzed using an Axioplan 2 microscope (Carl Zeiss MicroImaging, Inc. Thornwood, NY). Paraffin slides were dried at 60uC for 2 h before immunohistochemical staining. After deparaffinization, rehydration and depletion of endogenous peroxidase activity as described [15], tissue was stained with specific antibodies against Ki-67 (1:3000, Abcam, Cambridge, MA) and ATF3 (1:100, Santa Cruz Biotechnology, Santa Cruz, CA). For immunofluorescence staining, slides were incubated with antibodies, followed by incubation with Cy3, or FITC-conjugated secondary Abs (Molecular Probes) as described [16,17].

Proliferation Analysis by Ki67 Staining
Cellular proliferation was analyzed by immunohistochemical staining of formalin-fixed tissues using a rabbit polyclonal anti-Ki67 antibody. Ten randomly chosen sections from 5 animals per group were used for analysis. The number of Ki67-labeled nuclei in total 5000 hematoxylin-stained prostate epithelial cells was quantified using an Axioplan 2 microscope.

Cell Culture and Transfection
LNCaP human prostate tumor cells were purchased from American Type Culture Collection (ATCC, Manassas, VA) and maintained in RPMI1640 media (Invitrogen, Carlsbad, CA) supplemented with 10% FBS and 1% Penicillin/Streptomycin at 37uC under 5% CO 2 . The hTERT-immortalized normal prostate epithelial cells (PrEC) cells were generously provided by Dr. William Hahn (Dana-Farber Cancer Institute, Boston, MA) [18] and maintained in PrEGM medium. Cells were grown to ,80% confluence, at which time they were transiently transfected with small interfering RNAs (siRNAs) or gene expressing constructs. To silence ATF3 expression, cells were transiently transfected with ATF3 ON-TARGETplus SMARTpool siRNAs as well as 4 independent ATF3 siRNAs duplexes (Dharmacon Inc., Chicago, IL) using Nucleofector (Amaxa, Walkersville, MD). For ATF3 overexpression, an ATF3 construct (or empty vector control) was used for transient transfection of LNCaP cells.

Measurement of Cholesterol Level
Tissues or cells were finely minced in PBS on ice, membranes were isolated as described [19], and cholesterol levels were determined after lipid extraction using the Infinity Cholesterol Liquid Stable Reagent (Thermo Electron Corp., Waltham, MA).

Flow Cytometry
LNCaP cells were incubated in control medium or cholesteroldepletion medium for 18 h.
After trypsinization, cells were resuspended in PBS containing 2% FBS, and fixed in absolute ethanol at 4uC for 1 h. Cells were resuspended in 50 mg/mL propidium iodide staining solution and subjected to flow cytometry using a FACSCalibur flow cytometer (BD Biosciences). The apoptotic population of cells (sub-G 0 /G 1 ) was calculated by using CellQuest Pro (BD Biosciences).

RNA Extraction
Total RNA was purified from prostate lobes or cultured cells using a Qiagen RNEasy tissue extraction kit (Qiagen Inc., Valencia, California). RNA concentration was measured using a Nanodrop ND-1000 spectrophotometer (Thermo Scientific, Willmington, DE).

Microarrays and Data Analysis
Reverse transcription of total RNA and subsequent steps for sample probe preparation, microarray hybridization, washing and scanning of microarrays were performed by following a standard Affymetrix protocol at the Dana-Farber Cancer Institute microarray core facility. Oligo-based Mouse Genome 430A 2.0 Arrays (Affymetrix, for mouse microarray) or Human Genome U133A 2.0 Array (NuGEN V2, for LNCaP microarray) were used, respectively. For the mouse array experiment, two biological repeats were performed, while for the human array experiment, single samples were performed across a time course. Data normalization and summarization of expression values for each probe were calculated using the gcRMA method. We used the annotations of probesets that were provided by the Affymetrix website. All raw microarray data discussed here are deposited and available at Gene Expression Omnibus (http://www.ncbi.nlm.nih. gov/geo/) under the accession number GSE25500.
For the LNCaP data set, we then identified the differentially expressed genes (DEGs) as the genes with absolute log2-foldchanges with respect to CDM/time 0.0.585 (i.e. 1.5 fold). For the Hyper diet data, we used the following integrative hypothesis testing method: 1) we performed two-tailed T-test and log2median-ratio tests; 2) false discovery rates (FDRs) were computed using all possible randomization experiments for the individual statistical tests using Storey's method [20]; 3) the FDRs from the individual statistical tests were combined to result in the overall FDRs using Stouffer's method [21]. The DEGs were selected as the ones whose overall FDRs were ,0.05 and also absolute values fold changes .0.585. To integrate differential expression patterns of the DEGs in mouse and human, we used the ontology information obtained from the Mouse Genome Informatics (MGI) database (ver. 4.4). We searched for Gene Ontology Biological Processes (GOBPs) and known pathways that were statistically enriched by the DEGs using the DAVID software [22].

Network Modeling
We constructed a hypothetical network model using the 265 DEGs altered in both the hypercholesterolemic diet and CDM conditions. Information on protein-protein and protein-DNA interactions was gathered from the Kyoto Encyclopedia of Genes and Genomes (KEGG) and NCBI databases. We generated an initial network using the first neighbors of the 265 DEGs based on the interaction data. The initial network was then reduced to generate a subnetwork by removing the first neighbors of the DEGs that do not contribute to connecting DEG nodes. Finally, the nodes were grouped based on functional similarities based on GOBPs. To identify key central regulators among the DEGs, we computed 'degree centrality' of each regulator in the network [23]. A DEG regulator with high degree of centrality is considered a critical component that can regulate biological processes in the network [24].

Reverse Transcription and PCR
Primers for RT-PCR were designed using Mac Vector and were ordered from Integrated DNA Technologies (Coralville, IA). The cDNA amplification was performed using Invitrogen Platinum Blue PCR Supermix (Carlsbad, CA).

Cell Lysis and Western Blot Analysis
Tissue or cells were finely minced in PBS on ice and whole cell lysates were prepared in SDS-containing lysis buffer (62.5 mM Tris pH 7.4, 10% glycerol, 1% SDS and protease inhibitor cocktail) or in NP-40-containing lysis buffer (1% Nonidet P-40, 50 mM Tris pH 7.4, 10 mM NaCl, 1 mM NaF, 5 mM MgCl 2 , 0.1 mM EDTA, 1 mM PMSF and protease inhibitor cocktail). After centrifugation at 12,0006g for 15 min to remove debris, protein concentration was determined by microBCA (Pierce/ Thermo Scientific). Equal amounts of the lysates (20 mg) were subjected to SDS-PAGE and transferred onto nitrocellulose membranes for western blot analysis. Blotting signals were normalized by reprobing blots with anti-b-actin antibody, followed by densitometry for quantification.

Promoter Luciferase Assay
Luciferase reporter constructs with ATF3 promoter, wild type cyclin D1 promoter, or ATF3 binding mutant cyclin D1 promoter constructs were used for transient transfection of LNCaP or PrEC. Luciferase activity was measured in cell lysates using a luciferase assay kit (Promega, Madison, WI). Total protein was used for normalization. All experiments were carried out in triplicate and repeated 3 times using different preparations of plasmids.

Cell Proliferation Assay
LNCaP or PrEC were plated onto 6-well culture plates at a density of 1610 3 cells per well in standard growth medium. The following day cells were serum-starved for 16 h followed by treatment with the indicated medium: SF (serum free), CDM (cholesterol-depleted medium) or control medium. Cell proliferation rate was determined by crystal violet staining. Briefly, cells were stained with crystal violet solution and quantified by dissolving stained cells in 10% acetic acid solution. Colorimetric analysis was performed by measuring absorbance at 570nm.

Statistical Analysis
P-values were calculated using unpaired Student's t-tests or twoway ANOVA for simple comparisons. P,0.05 is considered statistically significant.

Molecular Responses to Variations in Cholesterol Level
In order to investigate the nature of the response of the normal mouse prostate to chronic changes in circulating cholesterol levels, we made 7-8 wk old male C57BL/6 and severe combined immunodeficient (SCID) mice (which lack T and B cells) hypercholesterolemic (the ''Hyper'' condition), and compared them with mice maintained at normal cholesterol levels (the ''Normo'' condition) for 4 months using an isocaloric diet procedure developed by our group [7]. Circulating testosterone/ DHT, insulin levels, prostate size and prostate volume were not statistically different between the 2 groups (Hyper vs. Normo), and these diets did not cause weight gain/loss or liver dysfunction (data not shown). Histological examination of prostate tissue from immune intact C57BL/6 mice revealed extensive, dense inflammatory infiltrates in the periprostatic adipose tissue (Fig. S1), but not in prostate parenchyma, in the Normo group (not shown). In order to detect primarily parenchymal, instead of inflammatory, responses to changes in circulating cholesterol, we used SCID mice in subsequent RNA expression profiling experiments in which circulating cholesterol was manipulated.
SCID mice fed the hypercholesterolemic (Hyper) diet demonstrated ,3 fold increased proliferation in prostate tissues, compared to the Normo condition, based on Ki67 staining index (Fig. 1A). As expected, Hyper mice showed ,2.5 fold increase in serum cholesterol (Fig. 1B). Cholesterol level in the membrane fraction of mouse prostate tissues was also increased .1.5 fold in the high cholesterol cohort (Fig. 1C), indicating that elevated circulating cholesterol affects membrane lipid composition of prostate cells in vivo.
To characterize molecular responses to the elevation in circulating cholesterol, gene expression profiling of SCID mouse prostate tissue was performed using Affymetrix mouse 430A.2 microarrays. We identified 1815 differentially expressed genes (DEGs) with a false discovery rate (FDR),0.05. 877 DEGs were upregulated and 938 were downregulated in the ventral prostate (VP) lobes from the Hyper vs. Normo groups. GOBP and KEGG pathway enrichment analysis (Table S1) demonstrated that the upregulated genes were mainly involved in cellular processes related to cell proliferation, inflammation, and chemotaxis (p,0.05), while the downregulated genes were related to biosynthesis and cell adhesion, as well as protein folding, transport, localization, and degradation. Genes involved in inflammation/ chemotaxis and cell proliferation/apoptosis suppression were largely upregulated (Table S1), consistent with the increased proliferation observed by direct examination of tissue (Fig. 1A).
As an alternative approach to testing the effect of altered cholesterol availability on prostate cells, cholesterol levels were reduced in cultured human prostate cells using a published procedure employing acute exposure to cholesterol-depleted medium (hereafter referred to as CDM) [25]. In addition to using LDL-deficient serum, this medium includes an HMG-CoA reductase inhibitor to reduce endogenous cholesterol synthesis, along with a titrated amount of mevalonic acid to prevent depletion of non-sterol end products, e.g. isoprenoids. Treatment with CDM suppressed proliferation of both LNCaP prostate cancer cells ( Fig. 2A) and normal human prostate epithelial cells (PrEC) (Fig. 2B), consistent with the essential role of cholesterol in cell proliferation. Incubating LNCaP cells in CDM resulted in a <60% reduction in intracellular cholesterol (Fig. 2C), but no cytotoxicity and no evidence of apoptosis above baseline was detected (Fig. 2D).
Gene expression profiling of LNCaP cells exposed to CDM conditions for 0, 3 or 16 h was performed using Human Genome U133A 2.0 microarrays. We identified 217 and 2477 DEGs in the 3 and 16 h conditions, respectively (See Materials and Methods; Fig. S2A and S2B). Analysis of gene expression patterns revealed 8 distinct groups (Fig. S2B), with the largest groups of DEGs (groups 4 and 5) not differentially expressed until 16 h after initiation of the experiment, while the second largest groups (groups 1 and 8) were differentially expressed from 3 h and continued to 16 h. Functional enrichment analysis of DEGs (Table S2) shows that the up-regulated genes induced by CDM treatment were mainly involved in cellular processes related to the folding, transport, localization, and degradation of proteins (p,0.05), while the down-regulated genes were involved in processes related to cell cycle progression, cell proliferation, inflammation, and immune response, consistent with the finding that CDM medium suppressed rates of cell growth (Fig. 2A). The results in Tables S1 and S2 indicate that similar cellular processes were affected by hyper-and hypocholesterolemic (Hypo) conditions, but in opposite directions.

A Cholesterol-sensitive Network
The overlap in the gene expression response to hypercholesterolemia in the mouse prostate and to CDM in LNCaP cells prompted us to integrate both data sets. We identified 449 genes altered in the 2 conditions (Fig. S3). These 449 DEGs were categorized into 4 groups based on the expression patterns between Hyper mouse prostates and CDM-treated LNCaP cells. This analysis narrowed the genes of interest to 265 (Clusters 2 and 3, Fig. S3A), whose expression levels were significantly altered in opposing directions in the 2 experimental settings. Using this gene set, we then constructed a hypothetical cholesterol-sensitive network to obtain a deeper and broader understanding of the genes/proteins underlying the prostatic response to changes in cholesterol.
The network (Fig. 3A) includes several interesting modules (clusters of functionally related genes based on GOBPs), including modules related to carbohydrate metabolism, inflammation, cell cycle regulation, cell migration, chromosome remodeling, RNA splicing, vesicular transport and steroid biosynthesis. The node and node boundary colors in Fig. 3A   assess the validity of this network, we used RT-PCR to assay for differential expression of a total of 8 genes, which we considered representative of all functional categories, in the in vivo Hyper vs. Normo conditions or in the cull culture CDM vs. normal medium conditions. Two genes with a high degree of centrality (See Materials and Methods), ATF3 (activating transcription factor 3, a mammalian ATF/CREB family transcription factor) and a known ATF3 target, ID1 (DNA-binding protein inhibitor 1), were tested under both conditions (Figs. 3B and 3C). As predicted from the microarray data, hypercholesterolemia reduced the mRNA levels of ATF3 as well as ID1, and enhanced those of C3 (apolipoprotein C3), S100A (S100 calcium-binding protein A) and IFI44 (gammainterferon-inducible protein 44) (Fig. 3B). Also consistent with the microarray findings, in the cell culture setting mRNA levels of ATF3, ID1, ID2, CDK2 (cyclin-dependent kinase 2), and VEGFC (Vascular endothelial growth factor C) were up-(ATF3, ID1 and ID2) or down-regulated (CDK2 and VEGFC) in response to CDM treatment (Fig. 3C). These findings support the validity of the network model shown in Figure 3A.

ATF3: a Central Node in the Prostatic Response to Changing Cholesterol Levels
To identify one or more ''key'' regulators of the prostatic response to cholesterol, we performed the following analysis. We selected genes for regulatory proteins (e.g. transcription factors, signaling molecules) from the 265 consistently altered DEGs. For each regulator, we then tallied its degree of centrality, based on the number of known DEG interactors (See Materials and Methods). The goal was to identify the key regulators with the largest degree of centrality in the network. This analysis identified ATF3 as one such potential key regulator because it has known targets ICA1 (islet cell autoantigen 1), ID1, TCF12 (transcription factor 12), ATF3 has been reported to participate in both cell proliferation and immunity [26,27], and due to alternative splicing functions, can act as both an activator and repressor of gene transcription [28][29][30][31][32]. As a further test of the network model, we focused specifically on ATF3 and its sensitivity to cholesterol. RT-PCR analysis using three prostatic lobes (anterior prostate, AP; ventral prostate, VP and dorsolateral prostate, DLP) demonstrated that ATF3 gene expression levels were significantly decreased in the Hyper, compared to the Normo condition (Fig 4A). Consistent with these downward changes seen in hypercholesterolemia, expression of the ATF3 protein was induced in a time-dependent manner in both PrEC and LNCaP cells in response to cholesterol lowering in CDM (Figs. 4B and 4C). ATF3 was also acutely induced in response to rapid depletion of membrane cholesterol with methyl-b-cyclodextrin (b-CD), a cholesterol-absorbing reagent (Fig. 4B, right panel). ATF3 mRNA level was increased approximately 2.5-fold after 3 h of CDM treatment and this increase was retained for at least 30 h (Fig. 4D). Assessment of the responsiveness of the ATF3 promoter to cholesterol by luciferase reporter assay showed that CDM evoked ATF3 transcriptional activation in LNCaP (Fig. 4E) as well as PrEC (Fig. 4F), consistent with the mRNA and protein data.
These data show that ATF3 expression is regulated by cholesterol level both in normal immortalized prostate epithelial cells and in androgen-dependent prostate cancer cells. Moreover, the inverse correlation of ATF3 level and cell proliferation suggests that altered ATF3 may mediate the effect of cholesterol on regulating cell proliferation.

ATF3 Regulates Prostate Cell Proliferation
We next performed functional studies in LNCaP cells to assess the biological role of ATF3. Enforced expression of ATF3 inhibited cell proliferation, an effect that was reversed by ATF3targeted RNA interference (RNAi) (Fig. 6A). Conversely, knockdown of ATF3 using RNAi increased cell growth (Fig. 6B), suggesting that the endogenous protein is growth suppressive. ATF3 levels were low in growing cells without added stimuli, requiring CDM treatment for visualization (Fig. 6B, upper panel). To rule out the off-target effects of siRNAs, 4 independent ATF3 siRNAs were tested. Consistent with results using the ATF3 siRNA pool, cell proliferation was enhanced when ATF3 expression was depleted by the 4 individual ATF3 siRNAs (Fig. 6C).
Notably, the increased ATF3 level (.1.5 fold) in response to CDM coincided with reduced cyclin D1 expression (<0.5 fold compared to time 0) (Fig. 6D). Because cyclin D1 plays an essential role in the cell cycle transition from early to mid-G1 phases, we asked whether there is a functional link between ATF3 and cyclin D1 in the context of cell proliferation. Fetal bovine serum (10%) raised cyclin D1 expression as expected in a time-dependent manner in serum-starved control cells. ATF3 knockdown by RNAi increased cyclin D1 levels even in the absence of serum stimulation (Fig. 6E). Cyclin D1 promoter-driven expression of luciferase was also enhanced in response to serum, and was almost completely suppressed by ATF3 overexpression (Fig. 6F). To determine whether cyclin D1 mediates proliferative effects of ATF3 in response to cholesterol alterations, we compared the response of the wild type cyclin D1 (WT) promoter, and an identical promoter with the ATF3-binding site inactivated by mutation (MUT), after incubation in two control conditions (medium 6 FBS) or CDM. LNCaP cells were transiently transfected with an ATF3 expression construct, a vector alone control, or with the AFT3 expression construct + ATF3 siRNAs. Proliferation rate was measured at the indicated times. ATF3 expression levels were verified by immunoblot (upper panels). Data are plotted as cell proliferation (fold) vs. condition (Vec, ATF3, ATF3+siATF3) 6 SD (n = 3). (B-C) Knockdown of ATF3 increases cell proliferation. LNCaP cells were transiently transfected with siATF3 (or siCon) and the number of cells at day 0 (grey bars) & day 3 (black bars) were measured. Four independent ATF3 siRNAs (siATF3-1, -2, -3, and -4) were transiently transfected, and cell numbers were determined at day 3. Data are plotted as cell proliferation (fold) vs. condition 6 SD (n = 3). (D) Effect of cholesterol depletion on ATF3 and cyclin D1 expression. LNCaP cell were treated in serum free media (SF; grey bars) or with CDM (black bars) for 16 h and the level of ATF3 and cyclin D1 were determined. Data were normalized to b-actin from the same blots. Immunoblot data are representative of the immunoblot result used in densitometry. Data are plotted as expression level (fold) vs. condition 6 SD (n = 3). (E) ATF3 regulates cholesterol depletion-induced cyclin D1 expression (immunoblot analysis). LNCaP cells were transiently transfected with siATF3 (or siCon). After serum starvation for 16 h, cells were stimulated with 10% serum for the indicated times. Immunoblot analysis was performed to determine cyclin D1 expression in ATF3 deficient cells. (F) ATF3 regulates cyclin D1 expression (promoter reporter analysis). LNCaP cells were transfected with promoter construct of cyclin D1 containing a luciferase reporter and followed by additional incubation with 6 serum for 6 h. Data are plotted as promoter activation (fold) vs. condition 6 SD (n = 3). (G) Promoter activation of cyclin D1 upon cholesterol alteration requires ATF3 binding on promoter region (promoter reporter analysis). LNCaP cells were transfected with a luciferase construct of a wild type (WT) or an ATF3 binding site mutated cyclin D1 (MUT) promoter. Promoter activity was measured 6 h after treatment with various conditions (6FBS or CDM). Data are plotted as promoter activation (fold) vs. condition 6 SD (n = 4). All experiments were performed a minimum of 3 times. *p,0.05, **p,0.01 (Student's t-test). doi:10.1371/journal.pone.0039448.g006 Cyclin D1 promoter activation in response to cholesterol alteration was diminished when the ATF3 binding site was mutated (Fig. 6G). These results imply that ATF3 downregulation under conditions of hypercholesterolemia may activate cyclin D1 expression and lead to enhanced cell proliferation.

Influence of Cholesterol on ATF3 Expression and Inflammatory Infiltrates
In order to further investigate whether circulating cholesterol levels alter ATF3 expression in vivo in immune-intact animals, 7-8 wk old male C57BL/6 mice (n = 18/group) were maintained under Hypo, Normo or Hyper conditions for 4 months. There were no significant differences in weight, insulin level, circulating testosterone/DHT level, and no detectable liver dysfunction, in the 2 groups. The Hypo diet resulted in circulating cholesterol levels about 60% lower than the Hyper diet (p,0.05), and the 3 diet regimens resulted in 3 distinct average cholesterol levels ( Fig. 7A). To assess the effect of circulating cholesterol level on ATF3 expression, we measured ATF3 mRNA and protein levels in C57BL/6 mouse prostate tissue under Hypo, Normo or Hyper conditions. Consistent with the findings presented above, ATF3 mRNA (Fig. 7B) and protein ( Fig. 7C and 7D) were markedly reduced, seemingly in a step-wise manner, with increasing cholesterol.
Prostate tissue from Hypo mice showed evidence of substantially decreased Ki67 staining (Fig. 8A) and lymphomononuclear cell accumulation (Fig. 8B), in comparison to the Normo (not shown) and Hyper groups, indicating that cholesterol reduction may reduce prostatic inflammation. In the Normo and Hyper groups, inflammatory cell clusters (L) were found closely associated with nerve tracts (N) and blood vessels (yellow arrows) in Hyper animals ( Fig. 8C(i)). The presence of lymphomononuclear infiltrates was negligible in the Hypo condition (data not shown). IHC analysis using marker proteins of immune cells (CD45, a general marker for lymphocytes; CD3, T cell marker) revealed that the monocytoid cells in the infiltrates are a mixture of B and T cells, with a predominance of B lymphocytes (Fig. 8C(ii)).

Discussion
In this study, we sought to gain insight into the mechanism by which the normal prostate responds to changes in cholesterol level in the microenvironment. We took a systems approach of identifying a cholesterol-responsive gene set and constructing a signaling network. We then tested this network using a series of independent approaches. We used an isocaloric diet method that allows isolation of cholesterol level as a variable independent of energy effects, animal weight variation, level of visceral fat or changes in circulating androgen [7]. Direct examination of prostate tissue from mice kept under chronic hypercholesterolemia demonstrated increased cell proliferation and under hypocholesterolemic conditions showed suppressed inflammatory cell infiltration. These findings suggest that circulating cholesterol may trigger pathophysiological changes in the prostate, and that lowering cholesterol may inhibit such changes. In vitro cholesterol depletion experiments showed that cell proliferation is inhibited by insufficient cellular cholesterol. By integrating unbiased gene expression data from the in vivo and in vitro experiments, we identified a network of cholesterol-regulated genes in the prostate. Further analysis indicated that the transcription factor ATF3 is likely to be an important node in this network. To our knowledge, this is the first systematic study of the signaling mechanisms in the prostate affected by changes in cholesterol levels.
ATF3 is a 21kDa leucine zipper (bZIP) transcription factor that belongs to the ATF/CREB protein family. ATF3 mRNA levels are low or undetectable in most tissues under most conditions, but are induced by a variety of environmental cues, including cytokines, xenotoxic agents and physiological stresses [26]. ATF3 has been identified as a suppressor of innate immunity, viral immunity, and allergic inflammation [26,[34][35][36][37][38]. ATF3 suppresses LPS-induced expression of interleukin (IL)-6 and IL-12 in macrophages by working in conjunction and in opposition to Rel (an NFkB subunit), an activator of IL-6 and -12 genes [37]. ATF3 is also activated in the lungs of mice after allergen challenge, where it contributes to negative regulation of the expression of IL-4, IL-5, IL-13 and other CCL (b-chemokine ligand) chemokine genes that mediate immune cell infiltration to pulmonary tissue. Consequently, ATF3-deficient mice exhibit increased airway hypersensitivity including enhanced eosinophilia [36]. Together, those studies suggest that ATF3 regulates inflammatory responses.
As a transcriptional repressor of cyclin D1, IRS (insulin receptor substrate) 2, and ID1 expression in chondrocytes, mouse embryonic fibroblasts, mouse pancreatic beta cells, and HaCaT human skin keratinocyte cells, ATF3 functions as an antiproliferative factor [39][40][41][42][43]. In cancer cells, ATF3 has also been identified as an inhibitor of Ras-stimulated tumorigenesis [43]. Those studies suggest that ATF3 induction correlates with cell cycle arrest, growth inhibition, and/or apoptosis. In contrast, ATF3 overexpression enhanced transcription of FN (fibronectin)-1, TWIST (twist transcription factor)-1 and Slug in MCF10CA1 breast cancer cells [44], and led to increased proliferation of DU145 prostate cancer cells [45]. Moreover, targeting ATF3 for knockdown inhibits cell adhesion and invasion capability in HT29 human colon cancer cells [46]. ATF3 increased apoptosis in untransformed MCF10A mammary epithelial cells, while it inhibited apoptosis in MCF10A breast tumor cells [44]. These discrepancies in function suggest that the biological role of ATF3 is highly dependent on physiologic conditions. The role of ATF3 in malignancy cannot be generalized as either being oncogenic or tumor suppressing; it exhibits properties of both functional classes. This duality of function has been explained as a consequence of: (a) cellular context; (b) post-translational modifications; (c) interacting proteins; (d) sub-cellular compartmentalization; and/or (e) differential complex formation of ATF3 homo-or heterodimers with other bZip proteins (e.g. c-Jun, Jun B and Jun D).
Hypocholesterolemic mice exhibited substantially reduced levels of inflammatory infiltrates in the prostate and relatively high expression of ATF3. Similarly, ATF3 mRNA level was rapidly increased by depriving cells of cholesterol in vitro, an effect that was sustained for at least 30 h (Fig. 4C). These data could be evidence for suppression by high cholesterol of a feedback regulatory mechanism that inhibits prostatic inflammation, although the present data do not provide a mechanistic explanation for how altered ATF3 expression might affect immune cell infiltration. One possibility is that chronic hypercholesterolemia may activate prostatic cells or/and adipocytes, through regulation of ATF3 levels, to secrete chemo-attractants (e.g. cytokines and chemotactic adipokines) that recruit immune cells.
Notably, dense inflammatory infiltrates were found adjacent to nerves, suggesting the possibility that prolonged high cholesterol might lead to chronic pain. Recruitment of inflammatory cells into nerve fibers has been identified as a likely source of visceral pain in chronic pancreatitis, and painful ejaculation symptoms occur in 5-31% of men with LUTS linked to BPH [47,48]. In light of our present data, these observations, suggest the possibility that prolonged high cholesterol may promote symptoms associated with benign prostatic disease.
Our cholesterol sensitive network model revealed alterations in an inflammatory signature in response to variation of cholesterol level (Fig. 3A): Expression of S100A, C3 and IFI44 was increased in the Hyper group, and decreased in the Hypo condition (Fig. 3). Pathway analysis suggested that the immune/inflammationrelated genes were significantly enriched in the Hyper condition (Table S1), suggesting an inflammatory response to high cholesterol. Accumulating evidence has linked pathologic or premalignant changes in the prostate with chronic inflammation. In particular, most BPH tissues show evidence of a chronic inflammatory reaction: (a) only 23% of prostate biopsies from 284 BPH patients were free of infiltrating inflammatory cells [49], (b) the presence of inflammatory infiltrates in BPH tissues was associated with increased rates of disease progression and elevated risk of acute urinary retention (Medical Therapy of Prostatic Symptoms (MTOPS) study), (c) human BPH stromal cells isolated from surgical specimens express all of the toll-like receptors (TLRs), which are central mediators of the innate immune system [50]; and (d) a recent report of a BPH signaling network suggested activation of inflammation through the TGFb-Smad2/3 signaling pathway [51]. Chronic inflammatory infiltrates in BPH nodules are mainly composed of T cells and macrophages [52][53][54], which are recruited to the prostate by chemoattractants, including IL-6, IL-8 and IL-15. In particular, IL-8 is highly expressed in BPH specimens [55] and stimulates prostatic stromal and epithelial growth during BPH progression [56].
In summary, we have identified elements of a cholesterolsensitive network in the normal prostate, thereby demonstrating that the prostate senses and can mount a response to chronic changes in circulating cholesterol. Because cholesterol-lowering drugs of several types are in widespread use, further study of the effects of cholesterol on the prostate and on lower urinary tract function may uncover novel therapeutic strategies. Figure S1 Inflammation was observed in H&E staining from C57BL6 mice, but not in SCID mice, in normal diet condition.