An Integrative Genomics Approach for Associating Genome-Wide Association Studies Information With Localized and Metastatic Prostate Cancer Phenotypes

Abstract

High-throughput genotyping has enabled discovery of genetic variants associated with an increased risk of developing prostate cancer using genome-wide association studies (GWAS). The goal of this study was to associate GWAS information of patients with primary organ–confined and metastatic prostate cancer using gene expression data and to identify molecular networks and biological pathways enriched for genetic susceptibility variants involved in the 2 disease states. The analysis revealed gene signatures for the 2 disease states and a gene signature distinguishing the 2 patient groups. In addition, the analysis revealed molecular networks and biological pathways enriched for genetic susceptibility variants. The discovered pathways include the androgen, apoptosis, and insulinlike growth factor signaling pathways. This analysis established putative functional bridges between GWAS discoveries and the biological pathways involved in primary organ–confined and metastatic prostate cancer.

Introduction

Prostate cancer is the most common noncutaneous malignancy and the second leading cause of cancer-related death in men older than 40 years in the United States.¹ In 2015, an estimated 220 000 men were diagnosed with prostate cancer, and 27 540 of these died of the disease.¹ Although there has been significant progress in the treatment of prostate cancer, several challenges persist such as the means to match patients with targeted therapies and understanding the molecular mechanisms underlying genetic predisposition to tumor aggressiveness. Nearly all the mortalities from prostate cancer are due to metastatic disease, typically through tumors that evolve to be hormone refractory or castrate resistant.² Advances in microarray technology over the past decade have enabled molecular classification of subtypes of prostate cancer^3,4 and discovery of potential clinically actionable biomarkers.⁵ However, this approach has been unsuccessful in determining which genes are potential drivers of prostate cancer aggressiveness as opposed to being consequences of the disease state.

Over the past decade, considerable efforts have been directed at discovering the genetic variants associated with an increased risk of developing prostate cancer using genome-wide association studies (GWAS).^6,7 These studies have revealed the genetic variants, primarily single-nucleotide polymorphisms (SNPs) and genes associated with an increased risk of developing prostate cancer.^6,7 However, GWAS information has not been leveraged and integrated with gene expression data from primary organ–confined and metastatic prostate cancers to identify molecular signatures predictive of disease aggressiveness. A limited number of GWAS have reported genetic susceptibility variants associated with tumor aggressiveness,^8–16 prostate cancer progression and mortality,^17,18 and survival.¹⁹ However, most of the GWAS-identified loci are not phenotype specific, and their association with clinical phenotypes remains poorly understood. This limited progress must be viewed against the recognition that most GWAS on prostate cancer were not designed to identify genetic variants associated with tumor-specific clinical phenotypes as they were largely focused on men diagnosed with prostate cancer irrespective of the disease state. This knowledge gap has hampered translation of GWAS discoveries into clinically actionable biomarkers to improve human health.

Integrative genomics that combines GWAS information with gene expression data has the promise of associating genetic susceptibility variants and genes with the clinical phenotypes.⁷ Indeed, several groups including ours have reported integration of GWAS information with gene expression data in prostate cancer.^20–25 However, the association of GWAS discoveries with primary organ–confined and metastatic disease has not been reported. Similarly, discovery of molecular networks and biological pathways enriched for genetic susceptibility variants involved in the 2 disease states has not been reported. The objectives of this study were manifold: (1) to investigate whether genes containing genetic susceptibility variants associated with and increased risk of developing prostate cancer are associated with primary organ–confined and metastatic tumors, (2) to discover a gene signature enriched for genetic susceptibility variants which distinguishes the 2 patient groups, (3) to determine whether the genes containing genetic susceptibility variants are functionally related and involved in similar biological processes, and (4) to identify molecular networks and biological pathways enriched for genetic variants which drive the 2 clinical phenotypes. Throughout this investigation, we have defined SNPs identified from GWAS as genetic susceptibility variants and used the genes containing these genetic variants as the units of association.

Materials and Methods

Genetic susceptibility variants and associated genes

Genetic variants and genes used in this study were derived from publicly available data obtained from published reports on GWAS and the websites hosting supplementary data for the respective reports.^7,22 Data collection was based on the guidelines proposed by the Human Genome Epidemiology Network for systematic review of genetic associations.^26–30 The details pertaining collection of data used in this study have been reported in our previous published reports.^7,22 Here, we provide a brief but detailed description of the data used in this study.

We reviewed a total of 140 published reports on GWAS. The reports were screened by title, abstract, and full-text review to identify the studies meeting our eligibility criteria. After screening, 100 studies that met our eligibility criteria were selected and subjected to further detailed review. The exclusion criteria for the 40 studies included removal of studies with insufficient or incomplete information, reviews, studies reporting only intergenic regions, and studies with very small sample sizes (ie, studies containing <500 subjects in cases and controls). For the remaining 100 studies used in this study, they were considered eligible if they met the following criteria: (1) must have been based on a case-control study design using unrelated individuals, (2) publications must have been of full length and published in peer-reviewed journals or online in English language before July 2015, (3) prostate cancer must have been diagnosed by histological examination, (4) the sample sizes must be more than 500 for the cases and more than 500 for the controls, (5) the study must have provided sufficient information such that genotype frequencies for both prostate cancer and controls can be discerned without ambiguity, and (6) the studies must have used the appropriate and recommended statistical methods to infer the associations by taking into account the covariates and accounting for population structure.²⁶

We manually extracted the information from the 100 studies meeting our eligibility criteria and the accompanying Web sites containing supplementary data. The extracted information included SNP identification number (rs-ID); evidence of association as determined by the P value; a composite of strong (P ⩽ 10⁻⁶), moderate (P = 10⁻⁴–10⁻⁵), and weak association (P = 10⁻²–10⁻³); gene name; and associated chromosome position to which the SNPs map as determined by the dbSNP database³¹ and the Human Genome Nomenclature database.³² This search yielded more than 400 SNPs mapped to 172 genes from a population of more than 350 000 cases and more than 350 000 controls. Table SA (Supplementary Material) provides information about the genetic variants and references or published reports from which they were derived.

A concern with GWAS studies is publication bias (a bias that occurs when only the most significant SNPs are reported, also known as “the winner curse”). To address publication bias, we included genes containing genetic variants with strong, moderate, and weak associations. The premise is that the presence of genetic variants with strong, moderate, and weak associations in genes of similar biological functions is likely to give a degree of confidence that the associations are likely genuine and could potentially have functional impact on clinical phenotypes.

Gene expression data

We used publicly available gene expression data generated from tumor and control samples derived from the white women. The data were downloaded from the National Center for Biotechnology Information’s Gene Expression Omnibus under accession numbers GSE1431 and GSE6604.^33–35 The methods regarding experimental design, sample preparations, and data processing have been described by the data originators.^33–35 Briefly, the data included a total of 194 samples distributed as follows: primary organ–confined tumors, N = 88; metastatic tumors, N = 25; and cancer-free controls, N = 81. These tissues were collected between 2002 and 2004. The protocols used for sample collection have been fully described by the data originators.^33–35 All the data were processed on the Affymetrix platform using the Human GeneChip U95A and standard Affymetrix protocols. Expression data (average scaled difference values) were processed and normalized using the Affymetrix Microarray Analysis Software (MAS 5.0). The data were filtered out to remove spiked control genes. The final data matrix consisted of expression profiles of ~12 000 probes. The probes were mapped to gene names using the batch query in the NETAFX an Affymetrix database which contains gene symbols.

The U95A Chip represents about 9000 unique genes. Of the 172 genes containing genetic variants associated with an increased risk of developing prostate cancer, 89 genes were represented on the Chip. As a result of limited Chip capacity, some genes containing SNPs associated with an increased risk of developing prostate cancer were not represented on the Chip. Genes not represented on the Chip were not evaluated for gene expression, but instead were evaluated using gene ontology (GO) information,³⁶ network, and pathway analysis to assess their functional relationships with genes represented on the Chip as described in the “Data analysis” section.

Indeed, many data sources exist in the public domain, including data derived from serum samples and tissue microarrays. However, because gene expression can be tissue, time, and platform specific, and because of the lack of clinical information for many existing publicly available data sets, our analysis focused on using gene expression data generated from primary organ–confined and metastatic tumor samples generated using the Affymetrix platform and U95A Human Chip only, a limitation that we readily acknowledge.

Data analysis

We compared gene expression levels between patients diagnosed with primary organ–confined tumors and matched controls, between patients diagnosed with metastatic prostate cancer and matched control samples, and between patients with organ-confined tumors and patients with metastatic tumors. Significant differences in gene expression levels between cases and controls and between the 2 disease states were assessed using a permutation t test as implemented in Pomelo³⁷ and GenePattern³⁸ software packages. Due to small sample sizes for metastatic prostate cancer, we did not partition the data into test and validation sets as such an approach would lead to bias resulting from sampling errors. Instead, we used the leave-one-out cross-validation procedure as our prediction and validation model to identify genes with predictive power.³⁹ This approach has been used successfully in gene expression data analysis with limited sample size to eliminate bias.³⁹ We used the false discovery rate (FDR) procedure⁴⁰ to correct for multiple hypothesis testing. Genes were ranked based on the P values and the FDR, and highly significantly differentially expressed genes were selected for each comparison.

We performed hierarchical clustering using GenePattern³⁸ to identify genes with similar patterns of expression profiles for primary organ–confined and metastatic disease and a gene signature distinguishing the 2 disease states. Prior to clustering, gene expression data were normalized using the median normalization, standardized and centered.⁴¹ We performed GO analysis to gain insights about the molecular functions and biological processes in which the genes containing genetic variants and other genes not identified by GWAS are involved. We performed additional analysis using the Ingenuity Pathway Analysis (IPA)⁴² to identify molecular networks and biological pathways enriched for genetic variants.

Results

We investigated whether genes containing genetic susceptibility variants are associated with primary organ–confined and metastatic tumors and are differentially expressed between the 2 disease states. We further conducted investigations to identify molecular networks and biological pathways enriched for genetic susceptibility variants in each disease state. This section describes our findings.

Associating genes containing genetic variants with primary tumors

To investigate whether genes containing genetic susceptibility variants associated with prostate cancer are associated with primary organ–confined tumors, we compared gene expression levels between patients with primary organ–confined tumors and matched control samples. After correcting for multiple hypothesis testing, the analysis produced a signature of 53 significantly (P < .05) differentially expressed genes associated with primary organ–confined tumors. Thirty-two genes were highly significantly (P < 10⁻⁵) differentially expressed and predictive of primary organ–confined tumors (Table 1). The list included the genes TERT, BIK, POLR2E, KLK3, LILRA, CTBBP2, ZNF652, FSHR, PRPH, MSMB, LMTK2, and CCHCR1 containing SNPs strongly (P < 10⁻⁸) associated with an increased risk of developing prostate cancer (Table 1). Interestingly, the list included the genes PARK7, TNFSF10, RUVBL1, GLI2, and KRT8 containing genetic variants with moderate association and the genes MYC, IGF2, STAT3, TTLL1, HOXB13, KLK13, KLK2, KCNQ1, RAB14, ERG, VAMP8, PRKCI, INS, and TH containing SNPs with weak associations (Table 1). Most notably, the analysis produced 4 genes, TERT, KLK3, TNFS10, and MSMB, containing genetic susceptibility variants associated with tumor aggressiveness. A complete list of genes containing SNPs associated with an increased risk of developing prostate cancer is presented in Table S1 provided as supplementary data to this report. Additional analysis on the whole data set revealed a signature of 100 highly significantly (10⁻⁶) differentially expressed genes not identified by GWAS.

Table 1.

List of the top 32 most highly significantly differentially expressed genes containing SNPs associated with an increased risk of developing prostate cancer that were found to be associated with primary organ–confined prostate cancer.

Gene symbol	Chromosome position	SNP-ID	GWAS P value	GE P value	FDR
MYC	8q24	rs10090154	4.00 × 10−2	5.00 × 10−6	1.85 × 10−5
IGF2	11p15.5	rs7127900	3.06 × 10−2	5.00 × 10−6	1.85 × 10−5
KRT8	12q13.13	rs4919743	3.64 × 10−4	5.00 × 10−6	1.85 × 10−5
TERT	5p15.33	rs2242652	2.70 × 10−24	5.00 × 10−6	1.85 × 10−5
STAT3	17q21	rs744166	3.00 × 10−2	5.00 × 10−6	1.85 × 10−5
BIK	22q13.31	rs5759167	1.30 × 10−12	5.00 × 10−6	1.85 × 10−5
POLR2E	19p13.3	rs3787016	7.22 × 10−7	5.00 × 10−6	1.85 × 10−5
TTLL1	22q13.1	rs5759167	3.01 × 10−3	5.00 × 10−6	1.85 × 10−5
KLK3	19q13.41	rs2735839	6.45 × 10−37	5.00 × 10−6	1.85 × 10−5
LILRA3	19q13.4	rs103294	5.34 × 10−16	5.00 × 10−6	1.85 × 10−5
HOXB13	17q21.32	rs8556	6.00 × 10−3	5.00 × 10−6	1.85 × 10−5
KLK13	19q13.33	rs2736433	5.00 × 10−2	5.00 × 10−6	1.85 × 10−5
KLK2	19q13.33	rs2735839	9.00 × 10−3	5.00 × 10−6	1.85 × 10−5
KCNQ1	11p15.5	rs231362	1.00 × 10−2	5.00 × 10−6	1.85 × 10−5
MTNR1B	11q21-q22	rs10830963	3.00 × 10−2	5.00 × 10−6	1.85 × 10−5
GLI2	2q14	rs11122834	5.00 × 10−6	5.00 × 10−6	1.85 × 10−5
CTBP2	10q26.13	rs4962416	2.70 × 10−8	5.00 × 10−6	1.85 × 10−5
RAB14	9q32-q34.11	rs942152	2.00 × 10−3	5.00 × 10−6	1.85 × 10−5
ZNF652	17q21.32	rs7210100	3.00 × 10−13	5.00 × 10−6	1.85 × 10−5
FSHR	2p21-p16	rs2268363	5.00 × 10−8	5.00 × 10−6	1.85 × 10−5
ERG	21q22.3	rs2836370	1.00 × 10−2	5.00 × 10−6	1.85 × 10−5
RUVBL1	3q21	rs7641133	1.0 × 10−4	5.00 × 10−6	1.85 × 10−5
VAMP8	2p12-p11.2	RS10187424	5.00 × 10−2	5.00 × 10−6	1.85 × 10−5
PRPH	12q12-q13	rs10875943	6.9 × 10−12	5.00 × 10−6	1.85 × 10−5
PRKCI	3q26.3	rs4955720	7.00 × 10−3	1.00 × 10−5	3.56 × 10−5
INS	11p15.5	rs7127900	3.06 × 10−3	1.50 × 10−5	5.13 × 10−5
TNFSF10	3q26	rs3774315	7.34 × 10−5	2.00 × 10−5	6.59 × 10−5
MSMB	10q11.2	rs10993994	8.70 × 10−29	3.00 × 10−5	9.21 × 10−5
LMTK2	7q22.1	rs6465657	1.10 × 10−9	3.00 × 10−5	9.21 × 10−5
CCHCR1	6p21.3	rs130067	3.20 × 10−8	4.50 × 10−5	1.33 × 10−4
TH	11p15.5	rs7127900	3.06 × 10−3	8.50 × 10−5	2.44 × 10−4
PARK7	1p36.23	rs6703670	9.09 × 10−4	9.00 × 10−5	2.50 × 10−4

Abbreviations: FDR, false discovery rate; GWAS, genome-wide association studies; SNP, single-nucleotide polymorphism; GE, gene expression.

Associating genes containing genetic variants with metastatic tumors

To investigate the association between GWAS information and metastatic disease, we compared gene expression levels between patients with metastatic tumors and matched control samples. The analysis produced a signature of 49 genes significantly (P < .05) associated with metastatic disease. Twenty-nine genes were highly significantly (P < 10⁻⁵) associated with metastatic prostate cancer (Table 2). The list included the genes AR, MSMB, FERMT2, PDLIM5, POU5F1, WWOX, CTBP2, C9ORF3, ITGA6, EHBP1, KLK3, FSHR, IL16, and ZNF652 containing SNPs strongly associated with prostate cancer (Table 2). The list also included the genes PARK7, SLC19A2 containing SNPs with moderate association, and the genes VDR, PRKCI, CDKN2A, BMPR1A, CPNE3, TTLL1, PIK3R1, INS, AKR1C3, SKIL, TCF7L2, and IRS2 with weak associations (Table 2).

Table 2.

List of the top 29 most highly significantly differentially expressed genes containing SNPs associated with an increased risk of developing prostate cancer that were found to be associated with metastatic prostate cancer.

Gene	Chromosome	SNP-ID	GWAS P value	Unadj. P	FDR_indep
VDR	12q12-q14	rs7975128	2.00 × 10−2	1.00 × 10−7	1.00 × 10−7
AR	Xq12	rs5919432	1.20 × 10−8	1.00 × 10−7	1.00 × 10−7
PRKCI	3q26.3	rs4955720	7.00 × 10−3	1.00 × 10−7	1.00 × 10−7
CDKN2A	9p21	rs10811661	4.00 × 10−2	1.00 × 10−7	1.00 × 10−7
MSMB	10q11.2	rs10993994	8.70 × 10−29	1.00 × 10−7	1.00 × 10−7
FERMT2	14q22.1	rs8008270	1.78 × 10−14	1.00 × 10−7	1.00 × 10−7
PDLIM5	4q22	rs17021918	4.00 × 10−14	1.00 × 10−7	1.00 × 10−7
TNRC6B	22q13	rs7291691	5.61 × 10−5	1.00 × 10−7	1.00 × 10−7
BMPR1A	8q24	rs11597689	3.00 × 10−2	1.00 × 10−7	1.00 × 10−7
POU5F1	6p21.33	rs6983267	7.50 × 10−6	1.00 × 10−7	1.00 × 10−7
CPNE3	8q21	rs4961199	2.79 × 10−2	1.00 × 10−7	1.00 × 10−7
WWOX	16q23.3-q24.1	rs11150069	9.43 × 10−6	1.00 × 10−7	1.00 × 10−7
CTBP2	10q26.13	rs4962416	2.70 × 10−8	1.00 × 10−7	1.00 × 10−7
TTLL1	22q13.1	rs5759167	3.01 × 10−3	1.00 × 10−7	1.00 × 10−7
C9orf3	9q22	rs3802458	4.00 × 10−6	1.00 × 10−7	1.00 × 10−7
PIK3R1	5q13.1	rs13156223	4.00 × 10−2	2.00 × 10−7	9.00 × 10−7
INS	11p15.5	rs7127900	3.06 × 10−3	4.00 × 10−7	2.10 × 10−6
AKR1C3	10p15-p14	rs4881400	3.00 × 10−2	6.00 × 10−7	3.10 × 10−6
ITGA6	2q31.1	rs12621278	9.00 × 10−23	8.00 × 10−7	4.00 × 10−6
SKIL	3q26	RS10936632	5.00 × 10−2	3.10 × 10−6	1.36 × 10−5
EHBP1	2p15	rs721048	8.00 × 10−9	3.10 × 10−6	1.36 × 10−5
PARK7	1p36.23	rs6703670	9.09 × 10−4	3.60 × 10−6	1.51 × 10−5
KLK3	19q13.41	rs2735839	6.45 × 10−37	7.00 × 10−6	2.76 × 10−5
FSHR	2p21-p16	rs2268363	5.00 × 10−8	1.07 × 10−5	4.05 × 10−5
IL16	15q26.3	rs7175701	9.80 × 10−8	1.54 × 10−5	5.60 × 10−5
ZNF652	17q21.32	rs7210100	3.00 × 10−13	2.13 × 10−5	7.46 × 10−5
TCF7L2	10q25.3	rs7903146	9.00 × 10−3	2.27 × 10−5	7.66 × 10−5
SLC19A2	1q23.3	rs3765227	1.26 × 10−4	4.16 × 10−5	1.35 × 10−4
IRS2	13q34	rs7986346	6.00 × 10−3	7.17 × 10−5	2.25 × 10−4

Abbreviations: FDR, false discovery rate; GWAS, genome-wide association studies; SNP, single-nucleotide polymorphism.

Interestingly, the list included 5 genes, MSMB, AR, POU5F1, WWOX, and KLK3, containing genetic variants associated with aggressive prostate cancer. A complete list of genes containing SNPs associated with an increased risk of developing prostate cancer is presented in Table S2 provided as supplementary data to this report. To determine whether any of the associations discovered using gene expression overlap between the 2 disease states investigated, we examined their P values. Seven genes (PRKCI, MSMB, CTBP2, INS, PARK7, KLK3, and ZNF652) were strongly associated with both primary organ–confined and metastatic disease. Additional analysis on the whole data set produced a signature of 100 highly significantly (10⁻⁶) differentially expressed genes not identified by GWAS.

Differences in gene expression levels between primary and metastatic tumors

Some primary organ–confined tumors progress to metastatic disease. The molecular mechanisms underlying disease progression are poorly understood. To identify a signature of genes containing genetic susceptibility variants which distinguish patients with primary organ–confined tumors from patients with metastatic disease, we compared gene expression levels between the 2 patient groups. After correcting for multiple hypothesis testing, this analysis produced a signature of 56 significantly (P < 0.05) differentially expressed genes. Thirty genes were highly significantly differentially expressed (Table 3). The list included the genes MSMB, AR, C9ORF3, WWOX, FERMT2, BIK, ZNF652, CTBP2, PDLIM5, EHBP1, MMP7, POU5F1, LILRA3, and GLI2 containing SNPs strongly associated with the risk of developing prostate cancer (Table 3). Intriguingly, the list included the genes GHR, KRT8, and PARK7 containing SNPs with moderate association and genes BMPR1A, TTLL1, PRKCI, MYC, PIK3R1, RAB14, HOXB13, GADD45A, NCOA4, BMP5, VAMP8, IGF, and VDR containing SNPs with weak association (Table 3). Among the identified genes included MSMB, AR, WWOX, and KLK3 containing SNPs associated with aggressive tumors. A complete list of genes containing SNPs associated with an increased risk of developing prostate cancer is presented in Table S3 provided as supplementary data to this report. Further analysis on the whole data set produced a signature of 100 highly significantly (10⁻⁶) differentially expressed genes not identified by GWAS.

Table 3.

List of the top 30 most highly significantly differentially expressed genes between primary organ–confined prostate cancer and metastatic prostate cancer.

Gene name	Chromosome	SNP-ID	GWAS P value	Unadj. P	FDR_indep
BMPR1A	8q24	rs11597689	3.00 × 10−2	5.00 × 10−6	1.78 × 10−5
MSMB	10q11.2	rs10993994	8.70 × 10−29	5.00 × 10−6	1.78 × 10−5
TTLL1	22q13.1	rs5759167	3.01 × 10−3	5.00 × 10−6	1.78 × 10−5
AR	Xq12	rs5919432	1.20 × 10−8	5.00 × 10−6	1.78 × 10−5
C9orf3	9q22	rs3802458	4.00 × 10−6	5.00 × 10−6	1.78 × 10−5
PRKCI	3q26.3	rs4955720	7.00 × 10−3	5.00 × 10−6	1.78 × 10−5
MYC	8q24	rs10090154	4.00 × 10−2	5.00 × 10−6	1.78 × 10−5
WWOX	16q23.3-q24.1	rs11150069	9.43 × 10−6	5.00 × 10−6	1.78 × 10−5
PIK3R1	5q13.1	rs13156223	4.00 × 10−2	5.00 × 10−6	1.78 × 10−5
KRT8	12q13.13	rs4919743	3.64 × 10−4	5.00 × 10−6	1.78 × 10−5
FERMT2	14q22.1	rs8008270	1.78 × 10−14	5.00 × 10−6	1.78 × 10−5
BIK	22q13.31	rs5759167	1.30 × 10−12	5.00 × 10−6	1.78 × 10−5
ZNF652	17q21.32	rs7210100	3.00 × 10−13	5.00 × 10−6	1.78 × 10−5
CTBP2	10q26.13	rs4962416	2.70 × 10−8	5.00 × 10−6	1.78 × 10−5
PDLIM5	4q22	rs17021918	4.00 × 10−14	5.00 × 10−6	1.78 × 10−5
EHBP1	2p15	rs721048	8.00 × 10−9	5.00 × 10−6	1.78 × 10−5
PARK7	1p36.23	rs6703670	9.09 × 10−4	5.00 × 10−6	1.78 × 10−5
MMP7	11q21-q22	rs11568818	1.56 × 10−11	5.00 × 10−6	1.78 × 10−5
RAB14	9q32-q34.11	rs942152	2.00 × 10−3	5.00 × 10−6	1.78 × 10−5
HOXB13	17q21.32	rs8556	6.00 × 10−3	5.00 × 10−6	1.78 × 10−5
GADD45A	1p31.2	rs520820	2.00 × 10−2	5.00 × 10−6	1.78 × 10−5
NCOA4	10q11.2	rs10761581	2.00 × 10−3	5.00 × 10−6	1.78 × 10−5
BMP5	6p12.1	rs3734444	3.00 × 10−2	5.00 × 10−6	1.78 × 10−5
GHR	5p14-p12	rs2940919	3.77 × 10−4	5.00 × 10−6	1.78 × 10−5
VAMP8	2p12-p11.2	RS10187424	5.00 × 10−2	5.00 × 10−6	1.78 × 10−5
IGF2	11p15.5	rs7127900	3.06 × 10−3	1.00 × 10−5	3.30 × 10−5
POU5F1	6p21.33	rs6983267	7.50 × 10−6	1.00 × 10−5	3.30 × 10−5
LILRA3	19q13.4	rs103294	5.34 × 10−16	4.00 × 10−5	1.30 × 10−5
VDR	12q12-q14	rs7975128	2.00 × 10−2	6.00 × 10−5	1.80 × 10−5
GLI2	2q14	rs11122834	5.00 × 10−6	8.00 × 10−5	2.40 × 10−5

Abbreviations: FDR, false discovery rate; GWAS, genome-wide association studies; SNP, single-nucleotide polymorphism.

Patterns of gene expression profiles in primary and metastatic tumors

Having established the association between genes containing genetic susceptibility variants with clinical phenotypes, we performed hierarchical clustering and GO analysis. We sought to investigate whether the discovered GWAS genes have similar patterns of expression profiles and are functionally related to one another and to non-GWAS genes. This analysis was performed for each disease state and between the 2 disease states.

Figure 1A shows the patterns of gene expression profiles for the top 53 genes containing genetic susceptibility variants associated with primary organ–confined tumors. The analysis revealed similarities in patterns of gene expression profiles and functional relationships among genes containing genetic susceptibility variants (Figure 1A). We discovered a cluster of 46 genes upregulated in tumors; among them, genes KLK3, MSMB, TNFSF10, TERT, and KLK2 containing genetic variants are associated with tumor aggressiveness. Combined analysis of GWAS identified and non-GWAS genes revealed similarities in patterns of gene expression profiles between the 2 sets of genes (Figure 1B).

Figure 1.

(A) Patterns of expression profiles for the 53 genes containing genetic variants strongly associated with an increased risk of developing prostate cancer evaluated in primary organ–confined prostate cancer (PCa) and control or normal samples. Patients are represented in columns and genes in rows. Red and blue colors indicate upregulation and downregulation, respectively. (B) Patterns of expression profiles for the 53 genes containing genetic variants strongly associated with an increased risk of developing prostate cancer and the 100 non–genome-wide association studies genes evaluated in PCa prostate cancer and control or normal samples. Patients are represented in columns and genes in rows. Red and blue colors indicate upregulation and downregulation, respectively.

The patterns of gene expression profiles for the top 29 genes containing genetic susceptibility variants strongly associated with metastatic prostate cancer are presented in Figure 2A. The analysis revealed a set of 24 genes upregulated in metastatic disease; among them, the KLK3 gene is associated with tumor aggressiveness (Figure 2A). The remaining 5 genes (EHBP1, TCF7L2, GADD45A, MMP7, and NCOA4) were downregulated (Figure 2A). Overall, the genes containing genetic susceptibility variants had similar patterns of expression profiles and were functionally related (Figure 2A). Combined analysis of GWAS and non-GWAS genes revealed similarities in patterns of expression profiles between the 2 sets of genes (Figure 2B).

Figure 2.

(A) Patterns of expression profiles for the 29 genes containing genetic variants strongly associated with an increased risk of developing prostate cancer evaluated in metastatic prostate cancer (MPCa) and control or normal samples. Patients are represented in columns and genes in rows. Red and blue colors indicate upregulation and downregulation, respectively. (B) Patterns of expression profiles for the 29 genes containing genetic variants strongly associated with an increased risk of developing prostate cancer and 100 non–genome-wide association studies genes evaluated in MPCa and control or normal samples. Patients are represented in columns and genes in rows. Red and blue colors indicate upregulation and downregulation, respectively.

The patterns of expression profiles for the 34 genes containing genetic susceptibility variants distinguishing patients with metastatic disease from those with primary organ–confined tumors are presented in Figure 3A. The signature revealed 2 distinct clusters of genes. A cluster containing 9 genes (CTBP2, AR, PRKC1, ZNF652, PDLIMS, VDR, PIK3R1, POU5F1, and TNRC68) consistently upregulated in metastatic prostate tumors and downregulated in primary organ–confined tumors (Figure 3A). This cluster included the AR and POU5F1 genes which contain genetic variants associated with aggressive tumors. The other cluster included a set of 25 genes consistently downregulated in metastatic prostate tumors (Figure 3A). Among the genes downregulated in metastatic tumors were TERT, KLK2, MSMB, and WWOX containing genetic variants associated with tumor aggressiveness (Figure 3A). Further analysis combining GWAS and non-GWAS genes revealed similarities and functional relationships between the 2 sets of genes (Figure 3B).

Figure 3.

(A) Patterns of expression profiles for the 34 genes containing genetic variants strongly associated with an increased risk of developing prostate cancer evaluated in primary organ–confined prostate cancer (PCa) and metastatic prostate cancer (MPCa). Patients are represented in columns and genes in rows. Red and blue colors indicate upregulation and downregulation, respectively. (B) Patterns of expression profiles for the 34 genes containing genetic variants strongly associated with an increased risk of developing prostate cancer and 100 non–genome-wide association studies genes evaluated in PCa and MPCa. Patients are represented in columns and genes in rows. Red and blue colors indicate upregulation and downregulation, respectively.

Overall, the analysis revealed similarities in patterns of expression profiles and functional relationships among GWAS and non-GWAS genes in all the 3 cases evaluated. However, as expected, there was considerable variation in patterns of gene expression profiles. This can be explained in part by the diversity of populations and phenotypes from which GWAS discoveries were derived.

Association with molecular networks and biological pathways

To gain insights about the broader biological context in which the genetic susceptibility variants operate, we performed network and pathway analysis. We sought to identify molecular networks and biological pathways enriched for genetic susceptibility variants in each disease state. The results showing the molecular networks enriched for genetic susceptibility variants in primary organ–confined prostate cancer are presented in Figure 4. Network analysis revealed that the genes containing genetic susceptibility variants with strong associations—MSMB, KLK3, FGF10, AR, NKX3-1, TERT, POU5F1, RFX6, FOXP4, ITGA6, CTBP2, FSMR, NCOA4, and SQRDL (in red font)—are functionally related and interact with one another (Figure 4). The analysis also revealed that genes containing genetic variants with moderate-to-weak associations—PRPH, MYC, GLI2, RUVBL1, HOXB13, IGF2, ERG, KLK2, PRKCI, POLR2E, and KRT8 (in blue font)—are functionally related and interact with genes containing genetic variants with strong association (Figure 4). In addition, genes containing genetic susceptibility variants were found to interact with genes RELA, RIPK3, UBC, EWSR1, and ANKRD7 not reported in GWAS (Figure 4).

Figure 4.

Gene interaction networks based on the set of genes containing genetic variants associated with an increased risk of developing prostate cancer and non–genome-wide association studies (GWAS) genes found to be highly significantly differentially expressed between patients with primary organ–confined prostate cancer and controls or normal samples. Genes containing single-nucleotide polymorphisms (SNPs) with strong associations are marked in red fonts. Blue fonts indicate genes containing SNPs with weak-to-moderate association. Genes in black font are non-GWAS genes. Nodes indicate the genes and the edges indicate interactions based on functional relationships. Information on functional relationships of the genes and biological processes in which the genes are involved is provided in Supplementary Table S4.

Functional analysis of the genes in highly significant networks using GO information revealed sets of genes predicted to be involved in cellular development, cellular growth and proliferation, cancer, hereditary disorder, organism injury and abnormalities, tissue development, cellular development, cell death and survival, cell-mediated immune response, and cellular compromise. Pathway analysis revealed sets of genes involved in the androgen, androgen biosynthesis, growth hormone, and apoptosis signaling pathways. A complete list of genes containing genetic susceptibility variants and genes not identified by GWAS is presented in Table S7 provided as supplementary data to this report. Also presented in Table S4 is the information on molecular functions and biological processes in which the genes associated with primary organ–confined prostate cancer are involved.

The results showing the molecular networks enriched for genetic susceptibility variants for the metastatic disease are presented in Figure 5. The analysis revealed molecular networks enriched for genetic susceptibility variants (Figure 5). The networks included the genes NKX3-1, KLK3, FGF10, AR, MSMB, ITGA6, NCOA4, CTBP2, TERT, CASP3, and PIK6R1 containing genetic susceptibility variants strongly associated with an increased risk of developing prostate cancer. The network also included genes PARK7, TCF7L2, CDKN2A, IRS2, and VDR containing genetic variants with moderate-to-weak associations and non-GWAS genes (Figure 5). Functional analysis using GO information revealed sets of genes predicted to be involved in posttranslation modification, nucleic acid metabolism, small-molecule biochemistry, cancer, endocrine system development and function, cancer, and gene expression. A complete list of genes containing genetic susceptibility variants and genes not identified by GWAS along with the biological processes and molecular functions in which they are presented in Table S5 provided as supplementary data to this report. The genes AR, KLK3, NKX3.1, ITGA6, MSMB, CTBP2, and NCOA4 overlapped between the 2 disease states (Figures 4 and 5). Pathway analysis revealed biological pathways enriched for genetic susceptibility variants, including prostate cancer, AKT, P53, apoptosis, VDR, and AR signaling pathways.

Figure 5.

Gene interaction networks based of the set of genes containing genetic variants associated with an increased risk of developing prostate cancer and non–genome-wide association studies (GWAS) genes found to be highly significantly differentially expressed between patients with metastatic prostate cancer and controls or normal samples. Genes containing single-nucleotide polymorphisms (SNPs) with strong associations are marked in red fonts. Blue fonts indicate genes containing SNPs with weak-to-moderate association. Genes in black font are non-GWAS genes. Nodes indicate the genes and the edges indicate interactions based on functional relationships. Information on functional relationships of the genes and biological processes in which the genes are involved is provided in Supplementary Table S5.

As noted earlier in this report, some primary organ–confined prostate cancers progress to metastatic disease. To address this question, we performed additional networks and pathway analysis using a set of genes distinguishing the 2 disease states. We sought to identify molecular networks and biological pathways enriched for genetic susceptibility variants that are potential drivers of disease progression, and we performed network and pathway analyses using the set of genes distinguishing the 2 disease states. The analysis revealed molecular networks enriched for genetic susceptibility variants (Figure 6). The network included the genes CTBP2, NKX3.1, TERT, KLK3, ITGA6, FGF10, THADA, and SKIL containing genetic variants strongly associated with prostate cancer. Also identified were the genes IGF2, AR, HOXB13, PARK7, NCOA4, VDR, KRT8, and GADD45A containing SNPs with moderate-to-weak associations and the genes AKT, RELA, EGFR, ER, HDAC1, UBC, and ELAVL1 not identified in GWAS.

Figure 6.

Gene interaction networks based of the set of genes containing genetic variants associated with an increased risk of developing prostate cancer and non–genome-wide association studies (GWAS) genes found to be highly significantly differentially expressed between patients with primary organ–confined prostate cancer and metastatic prostate cancer. Genes containing single-nucleotide polymorphisms (SNPs) with strong associations are marked in red fonts. Blue fonts indicate genes containing SNPs with weak-to-moderate association. Genes in black font are non-GWAS genes. Nodes indicate the genes, and the edges indicate interactions based on functional relationships. Information on functional relationships of the genes and biological processes in which the genes are involved is provided in Supplementary Table S6.

Functional analysis of the genes in the networks revealed genes predicted to be involved in cellular growth and proliferation, cellular development, cellular movement, hair and skin development and function, organ morphology, cancer, organ development, molecular transport, developmental disorders, cellular growth and proliferation, cell death and survival, reproductive system development and function, cellular assembly and organization, and lipid metabolism. A complete list of novel and SNP-containing genes showing the biological processes and molecular functions in which they are involved in primary organ–confined and metastatic prostate cancer is presented in Table S6 provided as supplementary data to this report. Pathway analysis revealed the growth hormone and prostate cancer signaling pathways as the potential drivers of metastatic disease.

Discussion and Clinical Significance

We integrated GWAS information with gene expression data to identify gene signatures enriched for genetic susceptibility variants associated with primary organ–confined and metastatic tumors and distinguishing the 2 disease states. The results revealed gene signatures associated with and distinguishing the 2 disease states. Several studies have combined GWAS information with gene expression data.^20,22,23,43 However, this is the first study to report gene signatures enriched for genetic variants involved in primary organ–confined and metastatic prostate cancer and to distinguish the 2 disease states. The clinical significance of the results from this study can be summarized as follows:

1. Patient stratification. The discovery of gene signatures enriched for genetic susceptibility variants involved in primary organ–confined and metastatic tumors is significant. Because genetic variants regulate gene expression,^44,45 they could potentially be used to stratify patients to guide and or prioritize treatment options. For example, genetic variants (mapped to genes) rs2267437 (XRCC6), rs3815824 (MVP),^12,46 rs871135(POU5F1P1), 47 rs3774315 (TNFSF10), rs6497287 (HERC2),¹⁶ rs1571801 (DAP2IP),¹⁰ and rs2735839 (KLK3)^8,47 are associated with more aggressive prostate cancer. Thus, if confirmed or validated, these genetic variants and genes could be used to stratify and prioritize patients for treatment on the basis of potential disease aggressiveness.

2. Patient screening using PSA. Common genetic susceptibility variants could be used to improve the interpretation of serum PSA levels.^48,49 Literature reports estimate that 40% to 45% of the inter-individual variability in measured serum PSA concentrations can be explained by genetic factors,⁵⁰ Studies have shown that SNPs in or near the KLK3 gene which encodes PSA can influence serum PSA concentrations and subsequently affect the frequency of prostate cancer screening and detection,^48–51 the interpretation of serum PSA values,^48,49 and the performance of PSA as a screening too.^52,53 Although our study did not focus on correlating GWAS information with PSA levels, a GWAS study has shown that several genetic variants associated with an increased risk of developing prostate cancer are strongly associated with serum PSA.⁴⁸

3. Reduction of unnecessary biopsies. In a clinical setting, abnormal PSA values and digital rectal examinations determine the need for prostate biopsy.⁸ However, literature evidence suggests that only 30% to 40% of patients with abnormal PSA values or physical examination are routinely diagnosed with prostate cancer on transrectal ultrasound-guided biopsy.⁵⁴ Recent studies have shown that genetic susceptibility variants are predictive of prostate cancer diagnosis on biopsy in men of European ancestry^55,56 and different racial/ethnic populations.^57–59 Thus, if confirmed, the genetic variants discovered in this study could be used to reduce the number of potentially unnecessary biopsies.

4. Risk prediction. One of the challenges in prostate cancer is risk prediction. Family history and genetic variants are measures of genetic susceptibility to prostate cancer.⁸ Several studies have used genetic variants to calculate a genetic risk score (GRS) in different populations.⁸ Published reports have suggested that GRS is significantly better in discriminative ability than knowledge of family history.^60,61

The genetic variants and respective genes associated with tumor aggressiveness discovered in this study if confirmed could be used for calculating the GRS and improving risk prediction algorithms currently in use.

Limitations

This study provides insights about the global biological context in which genetic variants associated with an increased risk of developing prostate cancer operate in primary organ–confined and metastatic tumors. However, limitations of the study must be acknowledged. The GWAS information used in this study is biased toward men of European ancestry, and gene expression data used in this study were derived from the white population. Genetic susceptibility and level of prostate cancer aggressiveness vary among race and ethnic populations.^59,62–64 In addition, gene expression can vary and differ among populations,⁶⁵ and the data set used in this study is small. Larger studies are needed to evaluate perturbations and potential clinical utility of genes and variants associated with an increased risk of developing prostate cancer in different race and ethnic populations. In this study, we did not study allele-specific expression and the functional impact of genetic susceptibility variants on genes and pathways. However, several published reports have documented allele-specific expression^66–69 in human populations and more recently in prostate cancer.^70–72 Recently, Larson et al⁷⁰ reported a comprehensive evaluation of cis-regulatory variation in the human prostate transcriptome using gene-level allele-specific expression. More recently, Whitington et al⁷¹ elucidated the gene regulatory mechanisms underpinning prostate cancer susceptibility. Moreover, evidence from the literature indicates that genetic variants regulate and account for differences in gene expression among populations.^73,74

Conclusions

The results in this report demonstrate that integrative analysis combining GWAS information with gene expression data is a powerful approach for linking genetic predisposition to clinical phenotypes in prostate cancer. This exploratory study establishes putative functional bridges between GWAS discoveries and biological pathways in primary organ–confined and metastatic tumors. Validation of these genetic variants using large test and validation data will permit additional tools that can be incorporated into algorithms that can improve clinical practice and risk prediction in prostate cancer.