Dysregulated gene expression predicts tumor aggressiveness in African-American prostate cancer patients

Molecular mechanisms underlying the health disparity of prostate cancer (PCa) have not been fully determined. In this study, we applied bioinformatic approach to identify and validate dysregulated genes associated with tumor aggressiveness in African American (AA) compared to Caucasian American (CA) men with PCa. We retrieved and analyzed microarray data from 619 PCa patients, 412 AA and 207 CA, and we validated these genes in tumor tissues and cell lines by Real-Time PCR, Western blot, immunocytochemistry (ICC) and immunohistochemistry (IHC) analyses. We identified 362 differentially expressed genes in AA men and involved in regulating signaling pathways associated with tumor aggressiveness. In PCa tissues and cells, NKX3.1, APPL2, TPD52, LTC4S, ALDH1A3 and AMD1 transcripts were significantly upregulated (p < 0.05) compared to normal cells. IHC confirmed the overexpression of TPD52 (p = 0.0098) and LTC4S (p < 0.0005) in AA compared to CA men. ICC and Western blot analyses additionally corroborated this observation in PCa cells. These findings suggest that dysregulation of transcripts in PCa may drive the disparity of PCa outcomes and provide new insights into development of new therapeutic agents against aggressive tumors. More studies are warranted to investigate the clinical significance of these dysregulated genes in promoting the oncogenic pathways in AA men.

six biomarkers had displayed a differential expression pattern in AA men 22 , and five PCa-associated genes have shown to be more methylated in tumor tissues procured from AA patients 23 . Additional evidence revealed that remarkable changes have occurred in epigenetic hallmarks of tumor tissues and these molecular events can be used as prognostic markers for tailored treatment of PCa patients 24 . Although microarray-based analyses have been widely used to segregate non-malignant versus malignant, low versus high tumor stages, localized versus metastatic, hormone-naïve versus castrate-resistant PCa patients, responders versus non-responders to radioand chemotherapeutic agents, yet they do not have the ability to differentiate gene expression that can further stratify PCa patients based on their races and ethnicities.
In this study, we compared microarray data in PCa tissue specimens collected from 412 AA and 207 CA men to identify differentially expressed transcripts and their predicted signaling pathways contributing to the disparity outcomes among AA men. We then validated top listed differentially expressed genes by quantitative RT-PCR, ICC, Western blot and IHC analyses in Formalin-Fixed Paraffin-Embedded (FFPE) PCa tissue sections and cell lines established from PCa of AA and CA patients.

Results
Identification of differentially expressed genes in AA patients with PCa. We initiated our study by retrieving microarray data of 619 PCa patients; 412 AA and 207 CA collected from 11 data sets deposited in the Gene Expression Omnibus (GEO) database. After retrieving these data, we considered the most significant differentially expressed genes at a fold change of ≥2. Of those, 362 transcripts were differentially expressed in PCa of AA compared to CA men (Supplementary Table S1). From these listed genes, we selected the top 27 genes, which have a highly significant difference (p < 0.001) as shown in Table 1. The upregulated genes were KLK2, COX5A, AZGP1, AMD1, ALDH1A3, MSMB, TPD52, OAT, TIMP4, APLP2, SOCS2, CD24, NKX3-1, SOD1, LTC4S, ANXA1, ACTA2 and HIF1A whereas downregulated ones were F3, SHH, ADIPOQ, PTGDR, ALOX12, CNR1, FGF2, PTGES and LOX. Some of these transcripts have been reported to have a potential role in cancer cell growth 25 , progression [26][27][28] , and angiogenesis 29 . Differentially expressed genes are associated with different biological processes in PCa cells. The next question was how these transcripts contribute to PCa progression in AA men. First, we looked into top listed differentially expressed genes, which are involved in different biological processes to change cancer cells into more aggressive phenotypes. Our results showed that these dysregulated genes were associated with the regulation of cell proliferation, differentiation, motility, adhesion, migration, apoptosis, hormonal response, signal transduction, fatty acid synthesis and metabolism, protein transport and response to oxidative stress (Table 2). Moreover, these transcripts were localized at different cellular compartments to carry out their assigned cellular functions. Some of transcripts were localized in extracellular matrix, extravesicular bodies "exosomes" to regulate cell-cell communications, in lipid rafts, and in cytosol (Table S2). We demonstrated that these genes might be involved in turn on the oncogenic signaling to promote PCa progression and metastasis within favorable cellular compartments. This notion needs additional validation, and therefore we attempted further bioinformatic analyses to support these findings.
Dysregulated signaling pathways and their correlation to clinical outcomes. Perceptibly, our goal here was to dissect the different signaling pathways in which these dysregulated genes are involved. The humanmine.org bioinformatic software was used to identify dysregulated pathways ( Fig. 1). Differentially expressed genes whose fold change value is greater than cut-off value of 0.7 were used as input. Our results showed that these genes are involved in multiple pathways of cancer, prostate cancer, focal adhesion, lipid metabolism, constitutive PI3K/AKT signaling, EGFR, PDGF, FGFR, ERBB2/DAP12 and MAPK signaling pathways (depicted in Table S3). We further investigated the association of these genes with clinical outcomes in PCa patients including age at diagnosis, pathologic grading, residual tumor, number of lymph nodes, PSA level and Gleason score as shown in supplementary Table S4.

Validation of selected differentially expressed genes in PCa cells.
The critical step in our study was to validate the expression of these candidate genes on mRNA and protein levels. We initiated our experiments by examining the gene expression by qPCR analysis using a large panel of PCa cells established from PCa patients of known AA and CA origin. We utilized both immortalized, non-tumorigenic RWPE-1 (CA-origin) and primary non-tumorigenic RC77N/E (AA-origin) as prostate epithelial control cells, LNCaP, 22RV1, DU-145 and PC-3 as PCa cells of CA-origin, and MDA-PCa-2b, RC77T/E, E006-AA and E006-AA-ht as PCa cells of AA-origin. Data from qPCR analysis demonstrated that APPL2, AMD1, NKX3.1, LTC4S, and TPD52 were significantly upregulated (p < 0.001), ALDH1A3 was downregulated (p < 0.001) while OAT did not show any significant difference in PCa versus normal cells (Fig. 2). A statistical significant difference (p < 0.001) was observed for each of APPL2, AMD1, LTC4S, OAT, NKX3.1, ALDH1A3, and TPD52 (p ≤ 0.05) when PCa of AA-origin compared to PCa of CA-origin cells as shown in Fig. 2. Expression patterns of selected genes in PCa of AA and CA cell lines was confirmed on protein level by immunofluorescence and Western blot analyses for LTC4S, TPD52 and OAT (Fig. 3A,B). These proteins had different pattern of nuclear and cytoplasmic staining in AA and CA PCa cells. However, nuclear staining was mostly observed in E006AA cells ( Fig. 3A) but it needs further study to determine

Validation of selected differentially expressed genes in human PCa FFPE tissues. Considering
relative limitations of PCa cell lines, we examined the pattern of these transcripts in FFPE PCa tissues collected from 39 AA and CA patients. Before initiating this study, we stained these tissue sections with H&E followed by microscopic examination to determine the ratio of tumor to normal cells for each case and we only selected tissue blocks that contained more than 50% tumor cells. Our results revealed that transcripts of TPD52, NKX3.1, LTC4S, APPL2, ALDH1A3, and AMD1 were significantly upregulated (p < 0.05) in tissues procured from AA compared to CA PCa patients as illustrated in Fig. 4. However, OAT did not showed any significant differences. We then correlated dysregulated genes in tumor tissues with clinical outcomes in PCa patients. As shown in Table 3, the levels of gene expression (median ΔCT) were used to stratify PCa patients into two groups; low and a high expression groups. The percentage of positive cores was significantly elevated in the high expression group of APPL2, AMD1 and TPD52 (p < 0.05). Likewise, the percentage of tumor involvement in the prostate gland showed a significant elevation in high expression group of ALDH1A (p = 0.038) and APPL2 (p = 0.054) compared to its counterpart group. The high expression of ALDH, AMD1 and OAT was correlated with prostate volume. To this extent, we validated the data from the bioinformatic analysis in human FFPE PCa tissues on an mRNA level, however, the protein expression in these tissues are necessary to evaluate the expression of these candidate genes in PCa tissues. We stained PCa tissue sections collected from 56 AA and CA patients with antibodies raised against OAT, TPD52 and LTC4S. In accordance with above-mentioned data, TPD52 (p = 0.0098) and LTC4S (p < 0.0005) showed higher protein expression in AA versus CA tissue sections; however, there was no significant change observed in OAT expression (p = 0.15544) as shown in Fig. 5A,B.

Discussion
In this report, we established a sharp contrast between the expression pattern of AA and CA PCa patients by analyzing microarray data from GEO database. Gene ontology and bioinformatics analyses revealed that these genes have a potential role in PCa aggressiveness in AA men by altering cellular signaling pathways in favor of tumor cells. We were able to validate this race-based contrast in the expression pattern on RNA and protein levels in PCa FFPE tissues and cell lines. Our bioinformatic analysis identified unique genes associated with multiple biological processes and cellular trafficking in aggressive tumor cells. These include dysregulated genes, which contribute to the response to steroid hormones, fatty acid biosynthesis, regulation of cell proliferation, adhesion and motility, regulation of cell migration, and protein kinase signaling pathway. These genes including but not limited to NKX3.1, SHH, EGFR, HIF1A, CTNNB1, FASN and others. Prior studies suggested that the function of NKX3.1 is frequently lost in castrate-resistant PCa and associated with genomic instability and biochemical relapse-free when combined with c-MYC 30 . Using genetically engineered mouse model, NKX3.1-PTEN mutant mice developed androgen-independent aggressive tumors 31 . The second gene linked to aggressive PCa phenotype is the Sonic Hedgehog (SHH). SHH pathway is involved in PCa angiogenesis, metastasis and development of drug resistance 32 . Another evidence is that SHH-Gli1 axis is associated with transforming malignant PCa stem cells into metastatic-like cells 33 . In the same context, epidermal growth factor receptor (EGFR) is another dysregulated gene whose signaling pathway is well known to be involved in cell proliferation, migration, adhesion and its overexpression is correlated with poor prognosis 34 . Hypoxia-inducible factor 1 (HIF-1), as one of our candidates, facilitates tumor cells to adapt for hypoxic conditions by regulating genes associated with hormone-refractory progression, angiogenesis, metastasis, and therapeutic resistance 35 . Additionally, the expression of β-catenin was higher in PCa and associated with disease progression 36 . Indeed, adaptive metabolic pathways and their linked lipid rafts are important step in the process of metastasis. For instance, overexpression of fatty-acid synthase (FSAN) is associated with PCa progression and metastasis 37 . In response to steroid hormones, we previously reported that circulating estrogen and expression of ERβ were substantially higher in PCa tissues of AA men 38 .
In addition to the above-mentioned dysregulated genes, we reported other novel genes where their roles in PCa aggressiveness need more investigations. We validated top listed genes in PCa cells and found that APPL2, AMD1, ALDH1A3, LTC4S, OAT and TPD52 were upregulated in PCa of AA compared to CA cells. On tissue level, these genes were upregulated in FFPE tissues of AA and were significantly correlated with prostate volume, percentage of positive cores and percentage of tumor involvement. In this study, we identified TPD52, AMD1 and LTC4S in addition to other dysregulated genes as potential candidates that might be associated with PCa aggressiveness among AA men. Other previous studies have supported our findings of the strong link between these candidate genes and tumor aggressiveness. For example, tumor protein 52 (TPD52) is an oncogenic protein expressed in malignant tissues including PCa 27,39,40 . The overexpression of TPD52 in LNCaP cells induced cell growth, colonogenic growth, migration and Akt activity 41 . Equally important, overexpression of S-adenosylmethionine decarboxylase 1 (AMD1) promotes tumor growth by increasing biosynthesis of polyamines, and foci formation anchorage-independent cell growth 42 . Chronic inflammations in the prostate gland account for ~20% of carcinogenesis of PCa 43 , and these inflammatory responses predicting tumor aggressiveness and poor clinical outcomes 44,45 . Interestingly, the prostate gland luminal epithelial layer adjacent to infiltrating immune cells shows atrophic appearance 46   In the light of this evidence, we observed different cellular localization of OAT, LTC4S and TPD52 in PCa cells procured from AA and CA, which may imply a possible role in tumor aggressiveness in AA but it needs further studies. The strength of our study includes the bioinformatics analysis performed on a large number of PCa of AA patients followed by prediction of the oncogenic pathways of dysregulated genes, and their correlation with clinical outcomes in AA men. One of the limitations of the study is the use of tissue specimens collected from one cohort in validation steps; however, we validated these genes on RNA and protein levels in a number of FFPE PCa tissues and cells collected from AA and CA patients. Therefore, our findings are presenting molecular foundations by which we determined the clinical significance of these dysregulated genes in segregation of PCa patients according to their race and their association with poor clinical outcomes in AA men. More studies are warranted to investigate how these genes promote oncogenic signaling pathways and drive tumor cells towards aggressiveness in AA men. In conclusion, our findings suggest that dysregulation of transcripts in large number of PCa of AA compared to CA men may explain the aggressive behavior of PCa. Our data provide new insights into novel as well as known candidates involved in PCa disparity and might be of clinical significance as prognostic markers or therapeutic targets in AA men at advanced stages of the disease.

Materials and Methods
Data collection. Expression microarray data of 619 PCa patients were collected from 11 data sets in the GEO database including 412 AA and 207 CA patients. The raw gene expression counts were normalized by linear normalization according to the following equation: where x denotes the gene expression count, and r denotes the read per million.
To perform differential gene expression analysis for AA and CA, two-sample t-test with procedures in SAM (Significance Analysis of Microarrays) was applied 53 . To identify the differentially expressed genes for other clinical features, we collected the correlation analysis results from Broad Institute 54 , and adjusted the p-values using BH method based on the number of genes used in this study.
False discovery rates (FDR). Adjusted p-value ≤ 0.05 and fold change of ≥2 were used for reporting significantly differentially expressed genes (unless otherwise noted) to reduce the number of false positives. If possible, the Benjamini-Hochberg procedure was used to control for FDR and is reported in the results 55 . Benjamini-Hochberg is used in DESeq2 output by default. P-values that have been adjusted are denoted to as adjusted P-values.
Pathway analysis and visualization. The differentially regulated pathways were generated from humanmine.org using the differentially expressed genes identified in gene level analysis 56 . We selected the differentially regulated pathways with adjusted p-values of less than 0.05. For pathway visualization, Pathview package in R was adopted.   . qPCR was performed using SYBR Green master mix (Bio-Rad, Hercules, CA, USA) on a Bio-Rad CFX96 detection system. PCR products were run on agarose gel to assure the specificity of each primer. The list of primer sets used in this study was described in supplementary Table S5. The fold change of gene expression was calculated relative to β-actin and 5S rRNA by comparing Ct method as described 59 .
Western blot analysis. Western blot analysis was performed as previously described 60 . Briefly, about 20 µg whole protein lysate was loaded onto a 4-20% SDS-PAGE gel (Bio-Rad, Hercules, CA) under reducing conditions. The fractionated proteins were transferred onto a nitrocellulose membrane (Bio-Rad, Hercules, CA), which was subsequently blocked with 5% bovine serum albumin for 1 hour. The membranes were incubated overnight at 4 °C with antibodies raised against OAT, TPD52, and LTC4S (Biorbyt, San Francesco, CA). Anti-GAPDH was used as an internal protein loading control (Santa Cruz Biotechnology, Dallas, TX). The membranes were washed thoroughly in washing buffer and incubated with the proper secondary antibodies for 1 hours at room temperature. After another series of washing, the membranes were developed and visualized by Odyssey ® Fc Imager and C-Digit Blot Scanner (LI-COR, Lincoln, NE).
Immunofluorescence. Immunofluorescence was carried out as previously described 60 . PCa cells were cultured in chamber slides (Fisher Scientific, Hampton, NH), washed and fixed in 4% paraformaldehyde. After another series of washing, cells were permeabilized and blocked with 2% BSA in TBST buffer. Cells were incubated overnight at 4 °C with primary antibodies as indicated. Next, cells were incubated with Alexa Fluor ® 488 secondary antibody, then stained with 4′ 6′-diamindino-2-phenylindole (DAPI) and mounting medium (Vector Laboratories, Burlingame, CA). Images were acquired under Nikon D-ECLIPSE C1si spectral laser-scanning confocal (Nikon Instruments, Melville, NY).
Immunohistochemistry. Immunohistochemical (IHC) staining with anti-OAT, anti-TPD52 and anti-LTC4S antibodies (Biorbyt, San Francesco, CA) was performed according to our reported protocol 60 . Briefly, tissue sections were de-waxed in xylene and rehydrated in descending series of ethyl alcohol. Tissue slides were then heated in 0.01 M citrate buffer pH 6.0 (Newcomer Supply, Maddison, WI) for 20 min in a steam cooker. The sections were immersed in 3% hydrogen peroxide for 10 min to block endogenous peroxidase activity. The slides were incubated overnight with primary antibodies at 4 °C. Bound antibody was detected by avidin-biotin complex peroxidase method using an ABC Elite Kit (Vector, Burlingame, CA, USA) with 3,3′-diaminobenzidine (DAP) as a chromogen. Tissues were counterstained with Mayer's hematoxylin solution and lithium carbonate as a bluing agent (Newcomer Supply, Maddison, WI). The immunostaining signals were visualized and captured using Eclipse 80i microscope (Nikon Instruments, Melville, NY). The intensity of the developed staining was blindly assessed by a cytopathologist (ABS). The histoscore was calculated as we described 60 .
Statistical analysis. Data were presented as mean ± standard error of mean. Comparison between experimental and their control counterparts were performed by applying Mann-Whitney test and Welch-corrected unpaired t-test using GraphPad Prism 7.0 (GraphPad Software, Inc., La Jolla, CA). An adjusted p-value of less than 0.05 was considered significant.

Data Availability
The microarray data and other associated generated data of the current study are available on request.