Common sequence variants on 2p15 and Xp11.22 confer susceptibility to prostate cancer

Abstract

We conducted a genome-wide SNP association study on prostate cancer on over 23,000 Icelanders, followed by a replication study including over 15,500 individuals from Europe and the United States. Two newly identified variants were shown to be associated with prostate cancer: rs5945572 on Xp11.22 and rs721048 on 2p15 (odds ratios (OR) = 1.23 and 1.15; P = 3.9 × 10⁻¹³ and 7.7 × 10⁻⁹, respectively). The 2p15 variant shows a significantly stronger association with more aggressive, rather than less aggressive, forms of the disease.

Recently, we presented results from a genome-wide association (GWA) analysis of prostate cancer in which we used the Illumina Hap300 SNP chip to genotype Icelandic samples^1,2. In subsequent efforts to identify additional variants, we used two criteria to guide our follow-up analysis of the GWA data. First, several studies on prostate cancer risk in relatives of affected men have reported that brothers are at greater risk compared to fathers^3,4, thereby implicating an X-linked or recessive mode of inheritance. Second, identifying genetic factors that preferentially associate to aggressive prostate cancer is of utmost importance for improving treatment selection for early stage disease. Hence, we decided to fast track the variant from our GWA analysis (Supplementary Methods online) showing the strongest association among the Illumina Hap300 SNPs located on the X chromosome, as well as a sequence variant on 2p15 that was selected after evaluating variants showing a greater risk in more aggressive prostate cancer (Gleason ≥7 and/or T3 or higher and/or node positive and/or metastatic disease) than in less aggressive disease (Gleason <7 and T2 or lower).

Allele A of the SNP rs5945572 (rs5945572 A) showed the most significant P value of markers located on the X chromosome in a GWA analysis of 23,205 Icelandic samples, with an allelic specific OR of 1.21 (P = 3.36 × 10⁻⁴; Table 1). The SNP is located on Xp11.22, a region that has not been previously implicated in prostate cancer and that is correlated with neither the androgen receptor locus on Xq12 nor the HPCX locus on Xq27–28. On chromosome 2p15, allele A of rs2710646 (rs2710646 A) had an OR of 1.16 (P = 7.79 × 10⁻⁴; Table 2) for all Icelandic prostate cancer cases and an OR of 1.33 (P = 3.73 × 10⁻⁵) for individuals diagnosed with aggressive disease (Supplementary Table 1 online). By examining the Utah CEPH (CEU) HapMap data, we identified numerous SNPs that are substantially correlated with either of the two anchor SNPs, rs5945572 and rs2710646 (D′ ≥ 0.9 and r² > 0.4). From this list we selected several SNPs not present on the Illumina Hap300 chip, representing different degrees of correlation with the anchor SNPs, for further genotyping in 1,500 and 800 Icelandic cases and controls, respectively. None of the additional SNPs were found to be more significantly associated with the disease than the anchor SNPs (Supplementary Table 2 online).

Table 1.

Association results for rs5945572 on Xp11.22 and prostate cancer in Iceland, The Netherlands, Spain, Sweden and the United States

Study population (cases/controls)	Frequency
Variant (allele)	Cases	Controls	OR (95% c.i.)	P value
Iceland (1,833/21,372)a
rs5945572 (A)	0.414	0.368	1.21 (1.09 – 1.34)	3.36 × 10−4
Holland (991/2,021)
rs5945572 (A)	0.390	0.347	1.20 (1.04 – 1.39)	0.015
Spain (539/1,594)
rs5945572 (A)	0.432	0.364	1.33 (1.10 – 1.61)	3.20 × 10−3
Sweden (2,865/1,722)
rs5945572 (A)	0.421	0.379	1.19 (1.05 – 1.34)	4.84 × 10−3
Baltimore, Maryland (1,516/554)
rs5945572 (A)	0.397	0.334	1.31 (1.07 – 1.60)	8.39 × 10−3
Chicago, Illinois (656/503)
rs5945572 (A)	0.402	0.354	1.23 (0.98 – 1.54)	0.072
Nashville, Tennessee (526/612)
rs5945572 (A)	0.392	0.353	1.18 (0.93 – 1.50)	0.18
Rochester, Minnesota (1,128/500)b
rs5945572 (A)	0.381	0.306	1.40 (1.11 – 1.76)	4.39 ×10−3
All excluding Iceland (8,221/7,506)c
rs5945572 (A)	–	0.348	1.24 (1.16 – 1.33)	2.57 × 10−10
All combined (10,054/28,879)c
rs5945572 (A)	–	0.351	1.23 (1.16 – 1.30)	3.95 × 10−13

All P values shown are two-sided. Shown are the corresponding numbers of cases and controls (n), allelic frequencies of variants in affected and control individuals, the allelic odds ratio (OR) with 95% confidence interval (95% c.i.) and P values. For additional information on assays and frequencies, see Supplementary Tables 4 and 5 online.

^aResults presented for Iceland were adjusted for relatedness using a method of genomic control (see Supplementary Methods).

^bResults presented for the Rochester study group were adjusted for relatedness by applying a pedigree correction factor (see Supplementary Methods).

^cFor the combined study populations, the reported control frequency was the average, unweighted control frequency of the individual populations, whereas the OR and the P value were estimated using the Mantel-Haenszel model.

Table 2.

Association results for rs721048 on 2p15 and prostate cancer in Iceland, The Netherlands, Spain, Sweden and the United States

Study population (cases/controls)	Frequency
Variant (allele)	Cases	Controls	OR (95% c.i.)	P value
Iceland(1,854/21,064)a
rs2710646 (A)b	0.227	0.202	1.16 (1.06 – 1.26)	7.79 × 10−4
rs721048 (A)	0.229	0.204	1.16 (1.06 – 1.27)	9.22 × 10−4
The Netherlands (998/2,021)
rs721048 (A)	0.205	0.187	1.12 (0.98 – 1.28)	0.095
Spain (548/1,616)
rs721048 (A)	0.234	0.210	1.15 (0.98 – 1.35)	0.088
Sweden (2,849/1,731)
rs721048 (A)	0.201	0.186	1.10 (0.98 – 1.23)	0.090
Baltimore, Maryland (1,521/557)
rs721048 (A)	0.205	0.206	1.00 (0.46 – 2.20)	0.99
Chicago, Illinois (665/552)
rs721048 (A)	0.198	0.168	1.22 (1.00 – 1.49)	0.055
Nashville, Tennessee (526/612)
rs721048 (A)	0.240	0.175	1.49 (1.21 – 1.83)	1.39 × 10−4
Rochester, Minnesota (1,132/501)c
rs721048 (A)	0.224	0.184	1.28 (1.06 – 1.55)	0.012
All excluding Iceland (8,239/7,590)d
rs721048 (A)	–	0.188	1.15 (1.09 – 1.22)	2.23 × 10−6
All combined (10,093/28,654)d
rs721048 (A)	–	0.190	1.15 (1.10 – 1.21)	7.66 × 10−9

All P values shown are two-sided. Shown are the corresponding numbers of cases and controls (n), allelic frequencies of variants in affected and control individuals, the allelic odds ratio (OR) with 95% confidence interval (95% c.i.) and P values based on the multiplicative model. For additional information on assays and frequencies, see Supplementary Tables 4 and 5 online.

^aResults presented for Iceland were adjusted for relatedness using a method of genomic control (see Supplementary Methods).

^bThe SNPs rs2710646 and rs721048 are highly correlated (r² = 0.99), but rs2710646 failed in genotyping in some of the non-Icelandic groups, and results are therefore only presented for this marker from the Icelandic study group.

^cResults presented for the Rochester study group were adjusted for relatedness by applying a pedigree correction factor (see Supplementary Methods).

^dFor the combined study populations, the reported control frequency was the average, unweighted control frequency of the individual populations, whereas the OR and the P value were estimated using the Mantel-Haenszel model.

We proceeded to genotype rs5945572 and rs2710646 in seven prostate cancer study groups of European descent, with populations from The Netherlands, Spain, Sweden and the United States. However, as the TaqMan assay for rs2710646 on 2p15 failed in design, we replaced it with an assay for a fully correlated SNP, rs721048 (linkage disequilibrium (LD) characteristics between rs2710646 and rs721048 in Icelanders and in the four HapMap populations (CEU, CHB, JPT and YRI; Supplementary Table 3 online); D′ = 1; r² ≥ 0.99), which was used for genotyping in the replication study populations. When results from all seven case-control replication groups were combined, using all prostate cancer cases, they were highly significant for both SNPs with OR = 1.24 (P = 2.57 × 10⁻¹⁰) for rs5945572 A on Xp11, and OR = 1.15 (P = 2.23 ×10⁻⁶) for rs721048 A on 2p15 (Tables 1 and 2). When we combined these data with the Icelandic data, the signals at both loci achieved genome-wide significance: rs5945572 A on Xp11.22 had an OR = 1.23 (P = 3.95 ×10⁻¹³), and rs721048 A on 2p15 had an OR = 1.15 (P = 7.66 × 10⁻⁹) (Tables 1 and 2). Removing all females (n = 14,135) from the control group in the analysis of the combined groups yielded essentially identical results for both loci (OR = 1.23 and 1.15 for rs5945572 A and rs721048 A, respectively; Supplementary Methods). In all of the replication groups, the observed effect for the two loci was in the same direction as in the Icelandic samples, except for rs721048 A, which in the Baltimore group showed no effect (OR = 1). However, a test of heterogeneity in the OR of the eight different study groups showed no significant difference for the two loci (P = 0.89 and 0.19 for Xp11 and 2p15, respectively). We note that in the results released by the Cancer Genetics Markers of Susceptibility study group⁵ (http://cgems.cancer.gov/data/), the two original anchor SNPs, rs5945572 and rs2710646, show nominal, but not genome-wide, significant association to prostate cancer, further supporting the data presented here.

For rs5945572 A on Xp11, the OR seen for cases with younger age at onset (≤65) or aggressive phenotype was the same as for the whole group. The frequency of rs721048 A, on the other hand, was significantly higher among individuals diagnosed with aggressive prostate cancer than among those with less aggressive disease (OR = 1.11; P = 2.6 × 10⁻³). Comparing the group of individuals with aggressive tumor (n = 4,787) to controls gave an allelic OR of 1.22 (P = 8.7 ×10⁻¹⁰) when combined for all study groups (Supplementary Table 1). Specifically, the heterozygous and homozygous carriers of rs721048 A, which are close to 31% and 4% of the general population, have a genotypic OR of 1.22 and 1.49 of being diagnosed with aggressive prostate cancer compared to the noncarriers, respectively. However, rs721048 A, like rs5945572 A, did not show a stronger association in individuals with younger age at onset of prostate cancer.

Both of the variants, rs5945572 on Xp11 and rs2710646 and rs721048 on 2p15, are located within regions characterized by extended linkage disequilibrium, according to the CEU HapMap results. On Xp11.22, the LD block spans the genomic region from 50.832 Mb to 51.553 Mb, covering 721 kb (NCBI Build 35). The 2p15 LD region is even larger—about 1.9 Mb (62.621–64.534 Mb; NCBI Build 35)—and contains several genes. Both of the 2p15 SNPs, rs2710646 and rs721048, are located within one of the introns of the EHBP1 gene. This gene is thought to be involved in endocytic trafficking and has not previously been implicated in cancer⁶. Through an RT-PCR analysis of various cDNA libraries, we detected expression of EHBP1 in several different tissue libraries, including those derived from the prostate (data not shown).

On Xp11, several genes have been localized to the 721-kb region of interest, but none have been previously linked with prostate cancer. Possible cancer candidate genes, based on previously described functions, are GSPT2 and MAGED1. GSPT2 is related to GSPT1, a gene encoding a GTP-binding protein that has an essential role at the G1-to S-phase transition of the cell cycle. MAGED1 has been implicated in programmed cell death through a JNK- and JUN-dependent mitochondrial pathway^7,8. The genes closest to the SNP rs5945572 are NUDT10 and NUDT11, along with a single exon transcript, LOC340602, with unknown function (University of California Santa Cruz Genome Browser, May 2004). NUDT10 and NUDT11 belong to a subgroup of phosphohydrolases that preferentially hydrolyze diphosphoinositol polyphosphates (DIPPS)⁹. It has been proposed that members of this protein superfamily may be involved in vesicle trafficking, stress responses, DNA repair and apoptosis^10,11. Using RT-PCR analysis of various cDNA libraries, we detected expression of GSPT2, MAGED1, LOC340602, NUDT10 and NUDT11 in several different tissue libraries, including those derived from the prostate (data not shown). Which one of these genes, if any, confers the risk identified by the association reported here remains to be shown.

Recently, several sequence variants have been identified that together account for a substantial proportion of the population attributable risk (PAR) of prostate cancer^1,2,5,12,13. With the identification of the two variants described here, yet another piece has been added to the puzzle of prostate cancer genetic susceptibility. Both the Xp11 and the 2p15 variants are common and confer a moderate risk, resulting in an estimated PAR of about 7% and 5% in individuals of European descent, respectively. However, the underlying causative biological perturbation linked to these variants remains to be elucidated.