Association Between Two Unlinked Loci at 8q24 and Prostate Cancer Risk Among European Americans

Abstract

Background

Recent studies have provided evidence of associations between genetic markers at human chromosome 8q24 and an increased risk of prostate cancer. We examined whether multiple independent risk variants exist in this region and whether the strength of observed associations differs as a function of disease aggressiveness.

Methods

We evaluated associations between 18 single-nucleotide polymorphisms (SNPs) in a 1-Mb interval at 8q24 and the risk of prostate cancer among 1563 case patients (1017 of whom had high-grade [Gleason score ≥7] and/or non–organ-confined disease) and 576 control subjects of European American ancestry. Differences in genotype frequencies between case and control subjects were compared using logistic regression analysis, with adjustment for age, and the Wald chi-square test. All statistical tests were two-sided.

Results

We identified multiple SNPs in a 50-kb region (referred to as locus 1) that are in linkage disequilibrium with a previously reported risk-associated SNP at 8q24, rs1447295, but were more strongly associated with prostate cancer risk in our study population. We also identified a novel susceptibility SNP, rs6983267, at a second locus (locus 2) that is approximately 70 kb centromeric of rs1447295 and in linkage equilibrium with, and independent of, locus 1. Risk alleles at locus 2 were common in our study population (minor allele frequency ∼50%, 25% homozygous for risk-associated allele). Analysis of the National Cancer Institute’s Cancer Genetic Markers of Susceptibility (CGEMS) prostate cancer association study database alone and in combination with our data provided further evidence for this second prostate cancer risk locus; in the combined analysis, the allele frequencies for rs6983267 differed statistically significantly between case patients and control subjects ( P = 1.61 × 10 ^-9 ). We also identified a third locus at 8q24, approximately 400 kb centromeric to locus 2, that was statistically significantly associated with prostate cancer risk in a combined analysis of our data and CGEMS study data ( P = 6.8 × 10 ^-4 ). A joint analysis of loci 1 and 2 indicated that 35% of the control subjects carried risk genotypes at one or both these loci; compared with men with the nonrisk genotype at both loci, men with risk genotypes at both loci had an odds ratio of prostate cancer of 2.68 (95% confidence interval [CI] = 1.62 to 4.43) and men with risk genotypes at either locus had an odds ratio of prostate cancer of 1.70 (95% CI = 1.39 to 2.07).

Conclusions

Three loci at 8q24 are independent genetic risk factors for prostate cancer.

CONTEXT AND CAVEATS

Prior knowledge

Various studies have linked genetic markers at human chromosome 8q24 with an increased risk of prostate cancer, but a comprehensive analysis of single-nucleotide polymorphisms throughout the region and the eventual identification of the causal variant(s) are still needed.

Study design

A case–control study to assess associations between germline variants at 8q24 and prostate cancer risk among men of European American ancestry.

Contribution

Three independent loci at 8q24 were identified that are associated in an additive fashion with an increased risk of prostate cancer.

Implications

The finding of multiple unlinked prostate cancer risk loci at 8q24 may provide important clues about the molecular mechanisms behind this genetic association.

Limitations

Control subjects were men with a serum prostate-specific antigen (PSA) level of 4.0 ng/mL or lower, potentially resulting in selection bias because control subjects (but not case patients) with certain diseases that are associated with higher PSA levels may have been preferentially excluded. Population stratification could have accounted for the observed different allele frequencies of variants between case patients and control subjects. There was no adjustment for multiple testing.

The evidence for a genetic susceptibility to prostate cancer has been well documented ( 1 ) and is some of the strongest among that for all common cancers ( 2 ). Multiple chromosomal regions have been identified as harboring major susceptibility genes for prostate cancer ( 3 ). Mutations in a number of genes located within these linkage regions have been identified in families with a history of prostate cancer; in some families those mutations cosegregate with prostate cancer ( 4–6 ). In addition, sequence variants in many candidate genes have been reported to be associated with prostate cancer risk in various case–control study populations ( 7 ). However, these prostate cancer linkage and association findings are, in general, inconsistent and replication studies have produced disparate results ( 8 ).

Amundadottir et al. ( 9 ) recently reported an association between markers at chromosome 8q24 and an increased risk of prostate cancer that was reproducible in multiple study populations. The 8q24 region was initially implicated in that study as a prostate cancer susceptibility locus by a genetic linkage analysis of prostate cancer families in Iceland. Subsequent analysis found that multiple alleles in the region, including at loci DG8S737 and rs1447295, were statistically significantly associated with an increased risk of prostate cancer risk in two case–control study populations from Iceland, and the association was confirmed in two additional case–control study populations of European ancestry. The association between these markers and prostate cancer risk was stronger among prostate cancer patients with high-grade tumors (i.e., Gleason score ≥7) than among patients with low-grade tumors (i.e., Gleason score ≤6), and approximately 8% of prostate cancers in the study population were estimated to be due to the association ( 9 ). In an independent report, Freedman et al. ( 10 ) confirmed the association between rs1447295 and prostate cancer risk (overall P <4.2 × 10 ^-9 ) by using four case–control study populations from the Multiethnic Cohort Study, including Japanese Americans, Native Hawaiians, Latino Americans, and European Americans.

Results from these two studies have provided strong support for an association between sequence variants at 8q24 and the risk of prostate cancer. However, a comprehensive analysis of single-nucleotide polymorphisms (SNPs) throughout the region and the eventual identification of the causal variant(s) are still needed. It is of particular interest to directly assess whether the 8q24 association extends to c-MYC and POU5F1P1, two genes known to reside in the 1-Mb region flanking rs1447295. c-MYC has been clearly implicated as a key gene in both human and experimental mouse prostate carcinogenesis ( 11 ). The putative protein encoded by POU5F1P1 is homologous to the stem cell marker octamer-binding transcription factor 4 encoded by POU5F1 ( 12 ) and deserves further evaluation in this respect.

In this case–control study, we comprehensively assessed associations between germline variants at 8q24 and prostate cancer risk among men of European American ancestry. The three major goals of this study were 1) to assess associations between prostate cancer risk and multiple SNPs that span a 1-Mb region at 8q24 that includes the c-MYC and POU5F1P1 genes, 2) to test for interactions of multiple SNPs with prostate cancer risk, and 3) to compare the associations of SNPs at 8q24 and prostate cancer risk between patients with more and less aggressive disease.

Methods

Study Subjects

Case patients were 1563 men of European descent (by self-report) who underwent treatment for prostate cancer at The Johns Hopkins Hospital from January 1, 1999, through December 31, 2006. Each patient's tumor was graded using the Gleason scoring system ( 13 ) and staged using the TNM (tumor–node–metastasis) system ( 14 ). We defined more aggressive and less aggressive disease based on tumor stage and Gleason score. Tumors with a Gleason score of 7 or higher or stage pT3 or higher or N+ or M1 (i.e., either high-grade or non–organ-confined disease) were defined as more aggressive disease. Tumors with a Gleason score of 6 or lower and stage pT2/N0 (i.e., cancer confined to the prostate) were defined as less aggressive disease. Normal seminal vesicle tissue that was obtained and frozen at the time of surgery was used to isolate DNA for genotyping of the case patients.

Men who were undergoing screening for prostate cancer at The Johns Hopkins Hospital and The Johns Hopkins University Applied Physics Laboratory (Columbia, MD) during the same time period were asked to participate as control subjects. Blood samples for preparation of DNA, serum prostate-specific antigen (PSA) levels, digital rectal examination results, and demographic information were available for these subjects. A total of 576 men met our inclusion criteria as control subjects for this study: European American ancestry by self-report, normal digital rectal examination, PSA level less than 4.0 ng/mL, and older than 55 years. The clinical and demographic information for these case patients and control subjects is summarized in Table 1 . The study received institutional approval and complied with Health Insurance Portability and Accountability Act regulations. Written informed consent was obtained from each participant.

Table 1

Clinical and demographic characteristics of prostate cancer patients and control subjects *

	Case patients
Characteristic	More aggressive disease †	Less aggressive disease	Control subjects
Number of subjects	1017	546	576
Mean age, y (SD)	60.1 (6.89)	56.8 (6.46)	59.6 (6.99)
Serum PSA level, No. (%)
≤4.0 ng/mL	76 (10)	197 (36)	576 (100)
>4.0 ng/mL	660 (90)	345 (64)	0 (0)
Pathologic stage ‡ , No. (%)
T2N0	186 (27)	528 (100)	N/A
pT3 or N1/N2	502 (73)	0 (0)	N/A
Gleason score, No. (%)
≤6	50 (6)	528 (100)	N/A
7	572 (70)	0 (0)	N/A
≥8	196 (24)	0 (0)	N/A

*

SD = standard deviation; PSA = prostate-specific antigen; N/A = not applicable.

†

For some patients with more aggressive disease, PSA values, stage, or grade information was not available.

‡

TNM (tumor–node–metastasis) system ( 14 ).

Selection of Single-Nucleotide Polymorphisms and Single-Nucleotide Polymorphism Genotyping

We selected 18 SNPs within a 1-Mb region at 8q24 (127.9–128.9 Mb) for genotype analysis. The selected SNPs included those that were most strongly associated with the risk of prostate cancer in the initial report of the 8q24–prostate cancer association ( 9 ) and in the prostate cancer genome-wide association data from the Cancer Genetic Markers of Susceptibility (CGEMS) project (phase 1A) ( 15 ), as well as those in regions containing the c-MYC and POU5F1P1 genes.

Polymerase chain reaction (PCR) and extension primers for the 18 SNPs were designed using MassARRAY Assay Design 3.0 software (Sequenom, Inc, San Diego, CA). The primer information is provided in Supplementary Table 1 (available online).

Genotyping procedures were performed as described in the iPLEX Application Guide (Sequenom Inc). All assay reagents were purchased from Sequenom except where indicated. Briefly, PCR assays were performed in a total volume of 5 μL that contained 10 ng of genomic DNA, 8.125 mM of MgCl ₂ , 0.5 U of HotStarTaq polymerase (QIAGEN Inc,Valencia, CA), 2.5 mM of each dNTP (Invitrogen, Carlsbad, CA), and 0.5 μM of each primer. The PCR cycling conditions were 94 °C for 15 minutes; followed by 45 cycles of 94 °C for 20 seconds, 56 °C for 30 seconds, and 72 °C for 1 minute; followed by a final extension at 72 °C for 3 minutes. Shrimp alkaline phosphatase (SAP) treatments were performed in a total volume of 7 μL that contained the entire PCR mixture and 0.3 U of SAP, with incubation at 37 °C for 40 minutes. The iPLEX extension reactions were performed in a total volume of 9 μL that included the entire SAP reaction and 1× iPLEX termination mix, 1× iPLEX enzyme, and 5.625 μM of each extension primer. The samples were desalted with 6 mg of clean resin and then dispensed to a SpectroCHIP. The chips were scanned with the use of a MALDI-TOF MS system, and the genotypes were analyzed by the MassARRAY Typer 3.4 (Sequenom Inc). All assays were performed by investigators who were blinded to the case–control status of the samples. The reactions were performed in 96-well plates. Each plate included the DNA from two reference subjects (1331-01 and 1331-02) from the Centre d’Etude du Polymorphisme Humain (CEPH, Paris, France) for quality control purposes and two wells in which water was substituted for DNA (negative controls). The average success rate for these 18 SNPs was greater than 98%, and the average concordance rate was greater than 99% for the two CEPH subjects across the plates.

Statistical Analyses

We used Fisher's exact test to test the Hardy–Weinberg equilibrium for each SNP separately among case patients and control subjects. Tests for pairwise linkage disequilibrium among these SNPs in control subjects (chi-square test, df = 1, two-sided), as well as estimates of D ′ and r² , were performed with the use of the SAS/Genetics software (version 9.0, http://support.sas.com/rnd/papers/sugi27/genetics.pdf ). Allele frequency differences between case patients and control subjects were tested for each SNP using a chi-square test with 1 df. Genotype frequency differences, assuming an additive, dominant, or recessive model, were also tested using unconditional logistic regression, with adjustment for age, and the Wald chi-square test. Odds ratios and 95% confidence intervals (CIs) for prostate cancer risk were estimated for men with risk genotypes compared with men with nonrisk genotypes under each of these genetic models. We tested for a two-locus interaction effect on prostate cancer risk by using an unconditional logistic regression method in which we sequentially fit locus 1, locus 2, and the product of locus 1 and 2 in the regression models. The best-fitting genetic models were used in the interaction analysis. Akaike information criterion (AIC) was used to obtain the most parsimonious model ( 16 ).

Associations between haplotypes of SNPs and prostate cancer risk were performed using a score test developed by Schaid et al. ( 17 ) as implemented in the Haplo.stat computer program ( http://mayoresearch.mayo.edu/mayo/research/biostat/schaid.cfm ). All statistical tests were two-sided, and a P value less than .05 was considered to be statistically significant.

Results

We genotyped 18 SNPs spanning 1 Mb at 8q24 in 576 control subjects and in 1563 prostate cancer case patients, of whom 1017 had more aggressive disease and 546 had less aggressive disease. The predominance of case patients with aggressive prostate cancer reflected our emphasis in this analysis on this clinically important category of disease. All 18 SNPs were polymorphic and had a minor allele frequency of at least 5% in the control subjects. Each of these SNPs was in Hardy–Weinberg equilibrium ( P ≥.05) in both case patients and control subjects.

Statistically significant differences ( P <.05) in allele frequencies between case patients and control subjects were found for multiple SNPs in this region ( Fig. 1, A ; Table 2 ). Specifically, four consecutive SNPs (rs1447295, rs4242382, rs7017300, and rs7837688) in a 50-kb region (hereafter referred to as locus 1) had statistically significant differences in allele frequencies between case patients and control subjects, with P values that ranged from 2.8 × 10 ^-4 to 1.12 × 10 ^-6 . These four SNPs were in strong linkage disequilibrium ( Table 3 ) and included rs1447295, the SNP that was found to be most strongly associated with prostate cancer risk in the original report by Amundadottir et al. ( 9 ). The frequency of the risk allele (A) of rs1447295 among the case patients and control subjects in this study was 13.2% and 7.9%, respectively. These estimates of allele frequencies were similar to those reported by Amundadottir et al. ( 9 ) for European Americans (12.7% in case patients and 8.1% in control subjects), by Freedman et al. ( 10 ) (13.1% in case patients and 10% in control subjects), and in the CGEMS database ( 15 ) (14.2% in case patients and 10.2% in control subjects). Haplotype analysis revealed a prostate cancer risk haplotype containing the risk alleles of these four SNPs that was present in 11.7% of case patients and 7.3% of control subjects (data not shown).

Table 2

Allele frequencies of 8q24 SNPs in case patients and control subjects *

			Allele frequency
SNP	SNP position	Risk allele	Case patients (n = 1563)	Control subjects (n = 576)	P†	Designated risk locus ‡
rs1467191	127951377	T	0.746	0.734	.45	–
rs10086908	128081119	T	0.738	0.712	.096	3
rs6981122	128163642	A	0.420	0.413	.71	–
rs16901979	128194098	A	0.055	0.045	.18	–
rs622556	128402379	T	0.497	0.477	.13	–
rs1668875	128410285	C	0.612	0.584	.096	–
rs6983267	128482487	G	0.568	0.509	6.12 × 10 −4	2
rs10505473	128487118	T	0.160	0.149	.4	–
rs12375310	128501388	C	0.695	0.687	.62	–
rs12334695	128523110	C	0.404	0.389	.4	–
rs1447295	128554220	A	0.132	0.079	1.90 × 10 −6	1
rs4242382	128586755	A	0.129	0.076	1.12 × 10 −6	1
rs7017300	128594450	C	0.154	0.110	2.80 × 10 −4	1
rs7837688	128608542	T	0.125	0.075	3.74 × 10 −6	1
rs7824074	128658004	G	0.701	0.706	.75	–
rs7841193	128746713	A	0.482	0.468	.42	–
rs4645959	128819722	A	0.963	0.958	.49	–
rs6470572	128837336	A	0.799	0.788	.43	–

*

SNP = single-nucleotide polymorphism.

†

Test for different allele frequencies between two groups using a chi-square test (two-sided test).

‡

Dashes indicate 8q24 SNPs that were genotyped and found not to be associated with prostate cancer risk.

Table 3

Pairwise test of linkage disequilibrium among 18 SNPs at 8q24*

Fig. 1

Allele tests for associations between single-nucleotide polymorphisms (SNPs) in a 1-Mb interval at 8q24 and prostate cancer risk. The statistical significance level [–log ₁₀ ( P )] of the allele test ( y -axis) is plotted against the position of the SNP ( x -axis). A ) Results from the Johns Hopkins study population. B ) Results from the Cancer Genetic Markers of Susceptibility (CGEMS) study population. C ) Pooled results for the Johns Hopkins and CGEMS study populations.

In addition to these four SNPs in locus 1, we also identified another SNP, rs6983267, located approximately 72 kb centromeric to rs1447295, for which the allele frequency also differed statistically significantly between case patients and control subjects ( P = 6.12 × 10 ^-4 ; Table 2 ). The frequency of the risk allele (G) of rs6983267 among case patients and control subjects was 56.8% and 50.9%, respectively. Because this SNP was in linkage equilibrium with the four SNPs in locus 1 (all pairwise D ′ < 0.18, r² = 0, P  ≥.05) ( Table 3 ), it represents an independent, novel prostate cancer risk locus at 8q24 (hereafter referred to as locus 2). This SNP is located 13318 bp upstream of the ATG start codon in the POU5F1P1 gene. SNPs within the c-MYC gene (rs4645959) or telomeric to the c-MYC gene (rs6470572) were not statistically significantly associated with prostate cancer risk.

To seek independent evidence for this second prostate cancer risk locus, we compared our results with those obtained in the genome-wide association scan, which are available through the CGEMS Web site ( 15 ) (https://caintegrator.nci.nih.gov/cgems/). The two strongest prostate cancer risk loci revealed in our study coincide with the two loci in this region that were most strongly associated with prostate cancer risk in the CGEMS database ( Fig. 1, B ), and the associations are in the same direction. For example, in the CGEMS data, multiple SNPs at locus 1 were associated with an increased risk of prostate cancer among 1182 prostate cancer patients and 1174 control subjects of European American ancestry, including rs1447295 ( P = 4.16 × 10 ^-4 ), rs4242382 ( P = 9.60 × 10 ^-5 ), rs7017300 ( P = 1.58 × 10 ^-4 ), and rs7837688 ( P = 3.8 × 10 ^-5 ). In addition, in the CGEMS data, multiple SNPs at locus 2 were also statistically significantly associated with prostate cancer risk, including rs6983267 ( P = 3.94 × 10 ^-4 ).

To assess the overall support for the 8q24 associations, we analyzed the pooled data for 2745 case patients and 1750 control subjects of European American ancestry from the CGEMS database ( 15 ) and from our study and found that these five SNPs in risk loci 1 and 2 were statistically significantly associated with prostate cancer risk ( Fig. 1, C ). Specifically, the allele frequencies for four SNPs at locus 1 differed statistically significantly between case patients and control subjects—rs1447295 ( P = 3.81 × 10 ^-9 ), rs4242382 ( P = 5.27 × 10 ^-10 ), rs7017300 ( P = 2.43 × 10 ^-7 ), and rs7837688 ( P = 5.88 × 10 ⁻¹⁰ )—as did SNP rs6983267 at locus 2 ( P = 1.61 × 10 ^-9 ). The combined analysis also identified another SNP that was statistically significantly associated with prostate cancer risk, rs10086908 ( P = 6.8 × 10 ^-4 ), which is located approximately 400 kb centromeric to locus 2 ( Fig. 1, C ). This SNP was statistically significantly associated with prostate cancer risk in the CGEMS database ( P = .007) but not in our study population ( P = .096); however, the direction of association was the same in both analyses. This SNP was in linkage equilibrium with SNPs in locus 1 and in locus 2 in our study (all pairwise D ′ < 0.23, r² = 0, P  ≥.05) ( Table 3 ). Therefore, rs10086908 represents another independent prostate cancer risk locus in the 8q24 region (hereafter referred to as locus 3).

We next performed association tests for each of these SNPs in our study population by assuming dominant and recessive genetic models ( Table 4 ; Supplementary Table 2, available online). For the four SNPs at locus 1, the strongest risk for prostate cancer was observed under a dominant model. The odds ratio for prostate cancer risk among men with the GT or TT genotype at rs7837688 versus men with the GG genotype was 1.90 (95% CI = 1.46 to 2.47; P = 1.3 × 10 ^-6 ). It is worth noting that approximately 14% of control subjects in our study (all of whom were of European American ancestry) carry the risk genotypes. The SNP at locus 2, however, was associated with the strongest risk of prostate cancer under a recessive model. The odds ratio for prostate cancer among men with the GG genotype at rs6983267 versus men with the TT or TG genotype was 1.43 (95% CI = 1.15 to 1.75; P = .0015). Approximately one-quarter of the control subjects in our study carry the GG risk genotype. The SNP at locus 3 was the most strongly associated with an increased risk of prostate cancer under a dominant model. The odds ratio for prostate cancer risk among men with the CT or TT genotype at rs10086908 versus men with the CC genotype was 1.43 (95% CI = 1.01 to 2.02; P = .04). Approximately half of the control subjects in our study carry the T allele of rs10086908.

Table 4

Associations between selected SNPs at 8q24 and prostate cancer risk *

Risk Locus	SNP	Genotype	No. of case patients (%)	No. of control subjects (%)	Model	P †	OR (95% CI)
3	rs10086908	CC	103 (6.64)	53 (9.25)
		CT	608 (39.20)	224 (39.09)	CT vs CC	.07	1.40 (0.97 to 2.01)
		TT	840 (54.16)	296 (51.66)	TT vs CC	.04	1.46 (1.02 to 2.09)
		CT or TT	1448 (93.36)	520 (90.75)	CT or TT vs CC (Dom)	.04	1.43 (1.01 to 2.03)
					TT vs. CC or CT (Rec)	.3	1.11 (0.91 to 1.34)
2	rs6983267	TT	285 (18.38)	132 (23.04)
		TG	771 (49.71)	299 (52.18)	TG vs TT	.16	1.19 (0.93 to 1.53)
		GG	495 (31.91)	142 (24.78)	GG vs TT	7.00 × 10 −4	1.62 (1.22 to 2.13)
		TG or GG	1266 (81.62)	441 (76.96)	TG or GG vs TT (Dom)	.02	1.33 (1.05 to 1.68)
					GG vs TT or TG (Rec)	1.45 × 10 −3	1.43 (1.15 to 1.75)
1	rs1447295	CC	1169 (75.61)	485 (84.94)
		CA	346 (22.38)	82 (14.36)	CA vs CC	2.55 × 10 −5	1.75 (1.35 to 2.28)
		AA	31 (2.01)	4 (0.70)	AA vs CC	.02	3.22 (1.13 to 9.16)
		CA or AA	377 (24.39)	86 (15.06)	CA or AA vs CC (Dom)	4.10 × 10 −6	1.82 (1.41 to 2.35)
					AA vs CC or CA (Rec)	.04	2.90 (1.02 to 8.25)
1	rs4242382	GG	1173 (75.73)	492 (85.71)
		GA	351 (22.66)	77 (13.41)	GA vs GG	1.64 × 10 −6	1.91 (1.46 to 2.50)
		AA	25 (2.52)	5 (0.87)	AA vs GG	.12	2.10 (0.80 to 5.51)
		GA or AA	376 (24.27)	82 (14.29)	GA or AA vs GG (Dom)	6.72 × 10 −7	1.92 (1.48 to 2.50)
					AA vs GG or GA (Rec)	.2	1.87 (0.71 to 4.90)
1	rs7017300	AA	1113 (71.81)	455 (79.27)
		AC	398 (25.68)	112 (19.51)	AC vs AA	1.88 × 10 −3	1.45 (1.15 to 1.84)
		CC	39 (2.52)	7 (1.22)	CC vs AA	.04	2.28 (1.01 to 5.13)
		AC or CC	437 (28.19)	119 (20.73)	AC or CC vs AA (Dom)	5.13 × 10 −4	1.50 (1.19 to 1.89)
					CC vs AA or AC (Rec)	.07	2.09 (0.93 to 4.70)
1	rs7837688	GG	1181 (76.24)	493 (85.89)
		GT	348 (22.47)	76 (13.24)	GT vs GG	1.86 × 10 −6	1.91 (1.46 to 2.50)
		TT	20 (1.29)	5 (0.87)	TT vs GG	.3	1.67 (0.62 to 4.47)
		GT or TT	368 (23.76)	81 (14.11)	GT or TT vs GG (Dom)	1.34 × 10 −6	1.90 (1.46 to 2.47)
					TT vs GG or GT (Rec)	.4	1.49 (0.56 to 3.98)

*

Sums less than 1563 and 576 for case patients and control subjects, respectively, reflect missing genotype data. SNP = single-nucleotide polymorphism; OR = odds ratio; CI = confidence interval; Dom = dominant; Rec = recessive.

†

Test for different genotype frequencies between two groups using a logistic regression analysis with adjustment for age (two-sided).

In our study population, the positive association between each of the four SNPs at locus 1 and prostate cancer risk was stronger in case patients with more aggressive disease than in case patients with less aggressive disease ( Table 5 ). For example, compared with men who had the GG genotype of rs7837688, men with the GT or TT genotypes had an odds ratio for prostate cancer of 2.03 (95% CI = 1.54 to 2.67; P = 3.2 × 10 ⁻⁷ ) when the analysis was restricted to patients with high-grade or high-stage disease; the odds ratio was 1.65 (95% CI = 1.21 to 2.26; P = .0017) when the analysis was restricted to patients with low-grade and low-stage disease. However, there was no statistically significant difference between the odds ratios for these two groups of patients for any of the four SNPs at locus 1. In addition, the four SNPs at locus 1 were also associated with younger age at diagnosis among case patients; for example, men who had the risk genotypes at rs1447295 were diagnosed an average of 0.77 year earlier than men with the nonrisk genotypes ( P = .03). This trend of stronger associations at this locus among patients with more aggressive disease was also observed in the CGEMS database ( 15 ). We could not test associations between the SNPs at locus 1 and age at diagnosis in the CGEMS study population because the individual data were not publicly available. By contrast to the SNPs at locus 1, the strengths of the associations between the SNPs at loci 2 and 3 and prostate cancer among case patients with more and less aggressive disease were inconsistent between the two studies; in our study, the associations were stronger, but not statistically significantly so, among case patients with less aggressive disease than among patients with more aggressive disease, whereas in the CGEMS study, they were stronger, but not statistically significantly so, among case patients with more aggressive disease than among patients with less aggressive disease.

Table 5

Association between selected SNPs at 8q24 and prostate cancer risk among case patients with or without aggressive disease *

			Allele frequency			Case patients with more aggressive disease vs control subjects				Case patients with less aggressive disease vs control subjects
Risk locus	SNP	Risk allele	Case patients with more aggressive disease	Case patients with less aggressive disease	Control subjects	P allele †	Model	P ‡	OR (95% CI)	P allele †	Model	P ‡	OR (95% CI)
3	rs10086908	T	0.737	0.739	0.712	.13	Dom	.14	1.31 (0.91 to 1.90)	.15	Dom	.02	1.73 (1.08 to 2.77)
2	rs6983267	G	0.563	0.577	0.509	.0033	Rec	.0081	1.37 (1.09 to 1.72)	.0012	Rec	.0012	1.54 (1.19 to 2.00)
1	rs1447295	A	0.137	0.122	0.079	8.51 × 10 -7	Dom	1.37 × 10 -6	1.93 (1.47 to 2.52)	.00078	Dom	.0022	1.62 (1.19 to 2.20)
1	rs4242382	A	0.135	0.118	0.076	3.66 × 10 -7	Dom	2.41 × 10 -7	2.03 (1.55 to 2.67)	.00084	Dom	.00065	1.71 (1.26 to 2.34)
1	rs7017300	C	0.163	0.136	0.110	4.33 × 10 -5	Dom	8.19 × 10 -5	1.62 (1.27 to 2.07)	.063	Dom	.089	1.28 (0.96 to 1.70)
1	rs7837688	T	0.132	0.112	0.075	8.02 × 10 -7	Dom	3.18 × 10 -7	2.03 (1.54 to 2.67)	.0031	Dom	.0017	1.65 (1.21 to 2.26)

*

SNP = single-nucleotide polymorphism; OR = odds ratio; CI = confidence interval; Dom = dominant; Rec = recessive.

†

Test for difference in allele frequencies between two groups using a chi-square test (two-sided test).

‡

Test for difference in genotype frequencies between two groups using a logistic regression analysis with adjustment for age (two-sided).

We also tested whether the statistically significant prostate cancer associations at locus 1 and locus 2 were independent in our study population. Because SNPs in these two loci were in linkage equilibrium, we used an unconditional logistic regression modeling method rather than haplotype analysis to perform the tests of association. The SNP rs7837688 was selected to represent all of the SNPs at locus 1 (the results of this analysis were the same when we used any of the SNPs in this linkage disequilibrium block; data not shown). We sequentially fit four different models: a model with locus 1 only (dominant), a model with locus 2 only (recessive), a model with both locus 1 and locus 2, and a model with both locus 1 and locus 2 as well as the product of locus 1 and locus 2. The interaction term (locus 1 × locus 2) in the last model was not statistically significant ( P = .98), indicating that these two loci had no multiplicative interaction effect on prostate cancer risk. The model that included both locus 1 (dominant) and locus 2 (recessive) was the most parsimonious model (i.e., it had the lowest AIC value; Supplementary Table 2, available online). Under this model, the odd ratio estimates for loci 1 and 2 were 1.89 (95% CI = 1.46 to 2.46; P = 2.01 × 10 ⁻⁶ ) and 1.43 (95% CI = 1.15 to 1.78; P = .001), respectively, and were essentially the same as the odds ratio estimates of 1.90 (95% CI = 1.46 to 2.47; P = 1.3 × 10 ⁻⁶ ) for rs7837688 at locus 1 and 1.43 (95% CI = 1.15 to 1.75; P = .0015) for rs6983267 at locus 2 from the single-locus analysis ( Table 4 ). These results suggest that these two loci are independent and measure different associations with prostate cancer.

We next estimated the magnitude of the additive effects of these two independent loci on prostate cancer risk in our study population ( Table 6 ). Subjects were grouped according to the following combinations of genotypes at rs7837688 (locus 1) and rs6983267 (locus 2): nonrisk genotype at both locus 1 and locus 2 (group 1), nonrisk genotype at locus 1 and risk genotype at locus 2 (group 2), risk genotype at locus 1 and nonrisk genotype at locus 2 (group 3), and risk genotype at both locus 1 and locus 2 (group 4). Compared with men with the nonrisk genotype at both loci (group 1), men with the risk genotype at locus 1 alone had an odds ratio for prostate cancer of 1.87 (95% CI = 1.38 to 2.53; P = 5.42 × 10 ⁻⁻⁵ ), men with the risk genotype at locus 2 alone had an odds ratio of 1.41 (95% CI = 1.11 to 1.79; P = .005), men with risk genotypes at both loci had an odds ratio of 2.68 (95% CI = 1.62 to 4.43; P = 1.00 × 10 ⁻⁴ ), and men with risk genotypes at either locus had an odds ratio of 1.70 (95% CI = 1.39 to 2.07; P = 1.48 × 10 ⁻⁷ ). It is important to note that the additive effect of these two loci on prostate cancer risk may have substantial public health implications. Among the men in our study population, approximately 3% carried the risk genotype at both risk loci and more than one-third carried a risk genotype at one of the two loci. The magnitude of the interaction between these two loci was similar among case patients with and without aggressive disease ( Table 6 ).

Table 6

Interaction effect of SNPs at locus 1 and locus 2 and prostate cancer risk *

Genotype combination (locus 1/locus 2)	Genotype		Number of subjects (%)
	rs7837688	rs6983267	Case patients (n = 1563)	Control subjects (n = 576)	OR (95% CI)	P†
All case patients
Nonrisk/nonrisk	GG	TT or GT	799 (51.61)	369 (64.40)	1.00 (referent)
Nonrisk/risk	GG	GG	381 (24.61)	123 (21.47)	1.41 (1.11 to 1.79)	.005
Risk/nonrisk	GT or TT	TT or GT	255 (16.47)	62 (10.82)	1.87 (1.38 to 2.53)	5.42 × 10 -5
Risk/risk	GT or TT	GG	113 (7.30)	19 (3.32)	2.68 (1.62 to 4.43)	1.05 × 10 -4
Risk allele at either locus	GT or TT	GG	749 (48.39)	204 (35.60)	1.70 (1.39 to 2.07)	1.48 × 10 -7
Case patients with more aggressive disease
Nonrisk/nonrisk	GG	TT or GT	528 (51.36)	369 (64.40)	1.00 (referent)
Nonrisk/risk	GG	GG	243 (23.64)	123 (21.47)	1.39 (1.08 to 1.80)	.01
Risk/nonrisk	GT or TT	TT or GT	181 (17.61)	62 (10.82)	2.05 (1.50 to 2.82)	9.02 × 10 -6
Risk/risk	GT or TT	GG	76 (7.39)	19 (3.32)	2.87 (1.70 to 4.83)	7.39 × 10 -5
Risk allele at either locus	GT or TT	GG	500 (48.64)	204 (35.60)	1.71 (1.39 to 2.11)	4.71 × 10 -7
Case patients with less aggressive disease
Nonrisk/nonrisk	GG	TT or GT	278 (51.67)	369 (64.40)	1.00 (referent)
Nonrisk/risk	GG	GG	142 (26.39)	123 (21.47)	1.41 (1.11 to 1.79)	.014
Risk/nonrisk	GT or TT	TT or GT	78 (14.50)	62 (10.82)	1.87 (1.38 to 2.53)	.017
Risk/risk	GT or TT	GG	40 (7.43)	19 (3.32)	2.68 (1.62 to 4.43)	.001
Risk allele at either locus	GT or TT	GG	260 (48.33)	204 (35.60)	1.70 (1.39 to 2.07)	1.723 × 10 -5

*

SNP = single-nucleotide polymorphism; OR = odds ratio; CI = confidence interval.

†

Test for difference in genotype frequencies between two groups using a logistic regression analysis with adjustment for age (two-sided).

The SNPs at the three prostate cancer risk loci identified in this study were not statistically significantly associated with serum PSA levels among control subjects (Supplementary Table 3, available online).

Discussion

The most important and novel finding of this study is the identification of multiple independent loci at 8q24 that are associated in an additive fashion with an increased risk of prostate cancer. Because the alleles associated with risk at these independent loci are common (i.e., at least one was present in more than one-third of the men in our study) and their effects are additive, our results indicate that 8q24 harbors factors that account for a substantial proportion of the genetic risk for this common disease. Our results, along with those of previous studies ( 9 , 10 ) and data in the CGEMS database ( 15 ), substantiate these loci at 8q24 as the genetic factors that are the most strongly associated with prostate cancer identified to date.

Our study population, which included a large number of case patients who underwent radical prostatectomy and for whom prostate tissue was available for accurate and reliable grading and staging, provided an excellent opportunity to assess associations between clinicopathologic tumor variables and genetic factors. In their analysis, Amundadottir et al. ( 9 ) found a statistically significantly stronger association between locus 1 and prostate cancer risk for case patients whose tumors had a Gleason score of 7 or higher than for patients whose tumors had a Gleason score of 2–6, although the difference in the magnitude of the associations was modest. By contrast, Freedman et al. ( 10 ) saw no such difference in association with tumor grade, although they compared different groups (i.e., Gleason score ≥8 versus Gleason score <8). In our study, the association with locus 1 was stronger among patients with more aggressive disease (94% of whom had tumor Gleason grades ≥7). A similar trend was noted in the CGEMS data ( 15 ). No consistent association with grade was found for the association between loci 2 or 3 and prostate cancer risk in our study or in the CGEMS data ( 15 ). Taken together, these results indicate that loci at 8q24 are associated with an increased risk for all prostate cancers. Larger studies may support a greater risk for more advanced disease.

The novel locus we identified, locus 2, is located approximately 70 kb centromeric of rs1447295 and 13 kb centromeric of the OCT4 pseudogene POU5F1P1. POU5F1P1 contains a poly-A tract at its 3′ end and is flanked by direct repeats, suggesting that its origin was via retrotransposition ( 18 ). However, it also contains a large open reading frame that potentially encodes a protein that is one amino acid shorter than, and 95% homologous to, wild-type Oct4 ( 18 ). Transcripts from POU5F1P1 have been found in various cancer cell lines ( 12 ), and another study ( 19 ) found that transcripts of OCT4 family members are present in both normal and cancerous prostate tissues. The OCT4 gene encodes a transcription factor that plays a critical role in maintaining the pluripotency of stem cells ( 20 ). Although a role of OCT4 in prostate cancer has not been investigated, current ideas about the contribution of stem cells to human carcinogenesis ( 21 ) suggest that its further study is warranted.

Our finding of multiple unlinked prostate cancer risk loci at 8q24 may provide important clues about the molecular mechanisms behind this genetic association. The multiple prostate cancer–associated sequence variants at 8q24 could affect the expression of some genes in the regions that flank the variants (cis effect) as well as genes in other regions of the genome (trans effect). Future efforts should focus on the identification of genes whose expression is influenced by genetic variants at 8q24, including c-MYC, POU5F1P1, and other genes throughout the entire genome.

Current prostate cancer diagnostic methods depend primarily on serum PSA levels to determine which men should undergo prostatic biopsy. Therefore, any factor that increases that serum PSA level could appear to be a prostate cancer risk factor even if it were independent of an effect on disease susceptibility, simply because it would prompt more biopsies. To examine whether such a possibility might explain the consistent observation of associations between loci at 8q24 and prostate cancer risk, we assessed the possible association between the SNPs at the loci characterized in this study and serum PSA levels among control men at time of screening. No association between these SNPs and serum PSA levels were found, suggesting that these SNPs are associated with prostate cancer risk directly rather than indirectly, e.g., as a result of increasing the rate of biopsy-driven diagnoses.

This study has several potential limitations. First, we limited the control subjects to men with a serum PSA level of 4.0 ng/mL or lower to minimize potential misclassification of case patients as control subjects. Although this approach most likely decreased the number of control subjects with undiagnosed prostate cancer and therefore increased our statistical power to detect associations with cancer risk, it is susceptible to selection bias because we may have preferentially excluded control subjects (but not case patients) with certain diseases that are associated with higher PSA levels, such as benign prostate hyperplasia. This potential selection bias could lead to false-positive associations for SNPs that may be associated with benign prostate hyperplasia but not with prostate cancer risk. Second, our study may be susceptible to population stratification; i.e., the observed different allele frequencies of SNPs between case patients and control subjects may be due to differences in their subregional origins (e.g., Northern or Southern Europe) rather than their disease status. However, it is unlikely that population stratification, if it existed in our study, could solely account for the associations we observed because the allele frequencies of rs1447295 and other SNPs in our study were similar to those reported in several other studies ( 9 , 10 ) and in the CGEMS database ( 15 ). Finally, we did not adjust for multiple testing in reporting the statistical significance levels of associations because many of these associations were not independent. However, multiple testing is not a major concern for this fine mapping study because the number of tests was relatively small. Furthermore, because of the highly statistically significant nominal P values for all the major findings reported in this study, all reported associations except for locus 3 would still be statistically significant even if we had used Bonferroni correction. Considering these potential limitations, we feel that our results should be interpreted with caution. Confirmation of our results in different study populations is warranted.

In summary, we have identified a novel prostate cancer risk variant at 8q24 that acts in an additive fashion with a previously identified adjacent locus tagged by SNP rs1447295. Because these loci are independent and the risk alleles at each are common, these loci together may account for substantially more prostate cancer in the population than previously appreciated.

Note added in Proof . After the submission of the manuscript, several papers describing studies identifying additional prostate cancer susceptibility loci at 8q24 were published, including one based on CGEMS data (Yeager et al., Genome-wide association study of prostate cancer identifies a second risk locus at 8q24. Nat Genet 2007;39:645–9), and studies by Haiman et al. (Multiple regions within 8q24 independently affect risk for prostate cancer. Nat Genet 2007;39:638–44) and Gudmundson et al. (Genome-wide association study identifies a second prostate cancer susceptibility variant at 8q24. Nat Genet. 2007;39:631–7).

Funding

National Cancer Institute (CA106523 and CA95052 to J. X., CA112517 and CA58236 to W. B. I., and CA86323 to A. W. P.); Department of Defense (PC051264 to J. X.).

The generous support of William T. Gerrard, Mario Duhon, Jennifer and John Chalsty is gratefully acknowledged.

S. L. Zheng and J. Sun contributed equally to the study.

The authors thank the study subjects who participated in this study. The study sponsor(s) had no role in the design of this study; the collection, analysis, or interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication.