Introduction

Genome-wide association studies (GWAS) have been instrumental in identifying novel regions of the genome that are associated with human diseases and traits (Donnelly 2008; Manolio et al. 2008). Though prostate cancer has a high incidence rate (~170 per 100,000 men in the United States) (Crawford 2003), until recently, the only well-established risk factors for prostate cancer were male sex, African ancestry, age, and family history (Johns and Houlston 2003). Genetic factors have been postulated to contribute to disease, and twin studies have shown high heritability (Lichtenstein et al. 2000). However, candidate gene association studies and early linkage analyses yielded inconclusive results. Recently, GWAS have identified at least 20 common genetic variants associated with prostate cancer risk (Amundadottir et al. 2006; Eeles et al. 2008; Freedman et al. 2006; Gudmundsson et al. 2007a, b; Haiman et al. 2007; Thomas et al. 2008; Yeager et al. 2007). These findings have forged new avenues for prostate cancer researchers to investigate regions that contribute to prostate cancer risk through yet to be defined mechanisms.

Two independent prostate cancer GWAS identified a single-nucleotide polymorphism (SNP; rs10993994 C>T) on chromosome 10q11.2, which maps to the proximal promoter of the microseminoprotein-β gene (MSMB) (Eeles et al. 2008; Thomas et al. 2008). The protein encoded by MSMB, prostate secretory protein 94 (PSP94), and its binding protein (PSPBP), have been evaluated as markers for early detection (Nam et al. 2006) and prognosis of prostate cancer (Bjartell et al. 2007; Reeves et al. 2006). The T allele of rs10993994 is significantly associated with the risk of prostate cancer (P = 9.7 × 10−19; OR = 1.20) (Lou et al. 2009), and follow-up work has shown evidence that alleles of rs10993994 differentially influence the expression of MSMB (Chang et al. 2009; Lou et al. 2009). While the C allele was shown to be required for MSMB expression, the T allele was associated with lower expression levels of MSMB in vitro (Lou et al. 2009). Interestingly, these results are consistent with the observation that MSMB expression levels progressively decrease during the development of prostate cancer from early to late stages (LaTulippe et al. 2002; Stanbrough et al. 2006; Vanaja et al. 2003). Together, these data provide a plausible mechanism for the contribution of the T allele to prostate cancer risk.

While there is strong evidence that variation at rs10993994 influences both MSMB expression and prostate cancer risk, it is possible that other genetic variants within this region of chromosome 10q14 could contribute to risk of developing prostate cancer, perhaps altering MSMB expression or through other mechanisms (Lou et al. 2009). Without a comprehensive assessment of common genetic variation across this region of 10q11.2, it is not possible to define the set of high priority common variants suitable for follow-up studies. In order to comprehensively catalog common genetic variation across MSMB and the neighboring region defined by the structure of the linkage disequilibrium, we used Roche-454 next-generation sequencing technology (Rothberg and Leamon 2008; Yeager et al. 2008) to resequence, in 70 unrelated individuals, a 97-kb region of chromosome 10 (chr10: 51,168,025–51,265,101) that includes a large block (51 kb) of linkage disequilibrium (LD) in which rs10993994 resides, MSMB, and a neighboring gene, nuclear receptor coactivator 4 (NCO4), that is also a possible candidate gene for prostate cancer risk.

Materials and methods

Samples

For sequence analysis, 38 prostate cancer cases and 27 controls were selected from the National Cancer Institute’s (NCI) prostate, lung, colorectal, and ovarian (PLCO) cancer screening trial (Gohagan et al. 2000); all (97%) but two were included in the initial prostate cancer GWAS (Yeager et al. 2007) that was previously conducted as a part of the Cancer GEnetics Markers of Susceptibility (CGEMS; see http://cgems.cancer.gov/) initiative. Additionally, 11 individuals from CEPH pedigrees were used in the analysis, including two sets of relatives. Seven of these individuals were CEU samples from The International HapMap Project (http://hapmap.org).

Region selection

Using CEU data obtained from the International HapMap Project (http://www.hapmap.org; The International HapMap Consortium 2005), based on the pattern of linkage disequilibrium (LD) flanking rs10993994 using Haploview (Barrett et al. 2005), we targeted 97 kb (chr10: 51,167,932–51,265,146), which included a 51-kb block of LD centromeric of MSMB with rs10993994 at its most telomeric position but none of the coding exons. We extended our target sequencing by an additional 46 kb to include MSMB and the telomeric neighboring gene, nuclear receptor coactivator 4 (NCOA4).

PCR primer design

Twenty sets of long-range PCR primers were designed to cover the 97-kb targeted region. Approximate amplicon size ranged from 3,544 to 5,542 bp; primer sets were designed to overlap, and averaged 292 bp overlap with the adjacent primer set. Primer design was as follows: primers had been initially designed using Primer3 (http://frodo.wi.mit.edu) (Rozen and Skaletsky 2000), were then quality-checked in silico for uniqueness, potential sequence paralogy and DNA repeat sequences using the BLAT feature of the UCSC genome browser (http://genome.ucsc.edu/cgi-bin/hgBlat). Next, NetPrimer (http://www.premierbiosoft.com/netprimer/index.html) was used to check for primer secondary structures and efficiencies. Primers were ordered from Integrated DNA Technologies (Coralville, IA, USA; http://www.idtdan.com). All primers, coordinates, and PCR conditions are supplied in Supplementary Table 1.

Sequencing

Post long-range PCR, all protocols were followed in accordance to standard kits used for the 454 GS FLX system (http://www.454.com/products-solutions/product-list.asp).

Polymorphism detection

An automated computational pipeline was developed to process sequence reads generated by 454 FLX genome sequencers. Whenever applicable, sequence reads from the same sample were pooled based on barcodes provided by Roche/454. QC was performed with vendor-supplied software, and sequence reads that passed QC were aligned to the target genomic region (chr10: 51,167,932–51,265,146) by MOSAIK software (http://bioinformatics.bc.edu/marthlab/Mosaik). The resulting assembly was analyzed in a column-by-column approach and potential polymorphic sites and most likely genotypes were called based on a set of heuristic rules. For each nucleotide position, the minimal sequence coverage depth was set to 20 reads. In addition, the ratio (r) of forward and reverse reads was determined. To avoid directional bias, an optimal range of r was set between 10 and 90%. Homozygous genotype calls were made when the most frequent allele was present in at least 85% of the reads. Heterozygous genotype calls were made when the two most frequent alleles were represented in 30–70% of reads. No genotype calls were made if the above criteria were not met. Manual inspections aided by the NextGENe software (http://www.softgenetics.com) and Consed (http://bozeman.mbt.washington.edu/consed/consed.html) were performed to QA the results and resolve ambiguous cases.

Descriptive statistics

Completion, concordance, minor allele frequency (MAF) estimations, deviations from fitness for Hardy–Weinberg proportion (HWP), pair-wise LD, and tag SNP selection were computed using the GLU software package (http://code.google.com/p/glu-genetics/).

Results

Coverage and depth

Excellent coverage and depth was obtained across the entire sequenced region. Figure 1 shows the coverage and depth averaged across all samples. The average depth was 104X, with a minimum of 24X; in two small regions (chr10: 51,177,997–51,183,539 and 51,241,521–51,246,917) that correspond to two amplicons we consistently observe an average coverage depth of <50X, though in both regions coverage did not fall below 25X.

Fig. 1
figure 1

Coverage and depth averaged over all samples for the 97-kb region of chromosome 10q11.2. The region consists of chr10: 51,167,932–51,265,146), which includes a 51-kb block of LD centromeric of MSMB with rs10993994 at its most telomeric position

Full size image

We consistently observed over-representation of sequence reads at the ends of amplicons, regardless of the genomic regions or individual samples sequenced. The coverage variability was as high as three or more times the average, an artifact that is most likely introduced during the shotgun sequencing library preparation of PCR amplicons (Harismendy and Frazer 2009). Another contributing factor to the “spikes” in Fig. 1 is that adjacent amplicons are usually overlapped at the ends, which increases the chance of sequence over-sampling.

Polymorphism detection and genotype and sample quality control

The heuristic rules we implemented (“Materials and methods”) took into consideration uneven sequence coverage. Therefore, coverage spikes at the amplicon ends (Fig. 1) have minimal impact on genotype accuracy. Genotypes were initially called for 685 SNPs and insertion/deletion polymorphisms (indels) in 76 individuals, including sites that had been previously reported in the NCBI dbSNP database (http://www.ncbi.nlm.nih.gov/projects/SNP/). Iterations of data cleaning excluded two related CEPH individuals and 41 loci with significant genotype issues (e.g., failure of Hardy–Weinberg proportions; P < 0.001) and/or with very low (<50%) genotype calling (completion) rates; 11 of these 41 polymorphisms fell within the chr10: 51,241,521–51,246,917 region of lower average coverage as discussed above. Within the other region of lower-than-average coverage, genotypes were successfully obtained for 18 loci, and the observed heterozygosity within this interval did not differ than the average across other regions. Of the remaining 637 loci, 190 sites in dbSNP were observed to be monomorphic within our samples and were excluded for subsequent analyses.

Utilizing the CGEMS prostate follow-up #2 iSelect (manuscripts in progress), we examined the 27 SNPs that overlapped the sequence data in 41 PLCO samples. In three samples, genotype concordance rates were <75%, all of which had low completion rates; all other samples were 100% concordant. After removing these three samples, genotype concordance per locus was 100% (Supplementary Table 2). Data from the International HapMap Project (http://www.hapmap.org) were used for 70 SNPs genotyped within this region over the seven HapMap CEU individuals. One sample displayed <95% concordance and was removed from further analyses. Concordance for two SNPs was problematic in the remaining six samples with all individuals displaying discordant genotypes; these two SNPs were excluded from further analyses. Three other SNPs displayed one discordant genotype each, and the remaining 65 SNPs displayed 100% concordance between HapMap genotypes and genotypes derived by sequencing.

Final dataset statistics

The final genotype dataset contained 70 individuals (36 and 26 PLCO prostate cancer cases and controls, respectively, and 8 unrelated CEPH individuals) and 440 polymorphic loci. The average call rate for genotypes was 92% (range 50–100%; median 97%; see Supplementary Table 2). Minor allele frequency (MAF) estimates were computed for each SNP and indel and overall averaged 15% (range 0.7–50%; see Supplementary Table 2); MAF by chromosomal position is shown in Fig. 2. Overall, 181 sites (169 SNPs and 12 indels) were observed to have a MAF ≥ 5% (Table 1). A total of 115 SNPs and 44 indels were observed in only one chromosome each. Since our indel-calling algorithm is currently still being refined, these low-frequency variants should be treated as preliminary; we include them here for completeness of probable polymorphic sites.

Fig. 2
figure 2

Minor allele frequencies by position for all polymorphic variants across the 97 kb of chromosome 10q11.2. All polymorphic sites are included across chr10: 51,167,932–51,265,146

Full size image
Table 1 Distribution of polymorphic SNPs and indels with regards to Illumina HumanHap610, dbSNP and HapMap inclusion
Full size table

Table 1 shows the classification of SNPs and indels with respect to previous dbSNP, HapMap, and Illumina HumanHap610 inclusion. Of the 440 polymorphic loci, 199 had previously been reported in the dbSNP database, only 84 of which contained available frequency information in dbSNP at the time of analysis (see Table 1). Within this region, 68 SNPs were genotyped as a part of the HapMap Project (http://www.hapmap.org), and 16 SNPs are included on the Illumina HumanHap610 fixed-content GWAS chip. In this study, we observed 241 novel SNPs and indels, of which 98 SNPs and 44 indels were observed as singletons and many others relatively rarely (<5% MAF).

No novel polymorphic sites were observed within the exons of the MSMB gene, nor were any of the exonic SNPs that had been previously reported (N = 24 in dbSNP) observed to be polymorphic within our samples. For NCOA4, no sites were observed to be polymorphic within the first six exons. However, within exons 7, 8, and 10 there were 2, 3, and 4 polymorphic sites observed, respectively. Supplementary Table 3 includes the subset of variants observed within the exons of the NCOA4 gene. Of the observed polymorphic sites, two were synonymous changes, tree non-synonymous, and four fell within the 3′UTR of the gene; all but two of the SNPs were observed in only one chromosome apiece (rs41306524, D223D; rs11548236, 3′UTR).

Linkage disequilibrium and tag SNP selection

An extremely high degree of LD was observed among polymorphic sites within the first 51 kb of the 97 kb sequenced region (Fig. 3; values shown are r 2). This region contains the previously associated functional SNP, rs10993994, as well as the proximal promoter of MSMB. Outside of this large block, significant LD was not observed within the region containing MSMB and NCOA4.

Fig. 3
figure 3

Linkage disequilibrium (LD) of polymorphisms ≥5% MAF as measured by r 2 across a 97-kb region of chromosome 10q11.2. Relative locations of rs10993994, MSMB, and NCOA4 are shown, though not to scale. Coordinates based on NCBI genome build 36.3

Full size image

Tagging analysis of the 97-kb region at an r 2 of 0.8 with rs10993994 as an obligate-include gave a total of 36 tags necessary to cover the 181 loci with a MAF > 5% in individuals of European background. The bin containing rs10993994 contains 20 other SNPs but no indels. Supplementary Table 4 shows the pairwise r 2 of these SNPs ranges from 0.8 to 0.85. It is notable that no other variant was observed to be completely correlated with rs10993994. Tagging at more aggressive r 2 thresholds of 0.9 and 1.0 increased the number of tags to 39 and 116, respectively. Supplementary Table 5 lists the bin and tag SNP information using three thresholds: r 2 ≥ 0.8, r 2 ≥ 0.9, and r 2 = 1.0.

In order to determine the overall contribution of the newly described variants from this study with respect to coverage of the 97 kb region, we tagged (r 2 ≥ 0.8, minimum MAF = 5%) based on prior inclusion of SNPs and indels in dbSNP, both (1) with and (2) without reported frequency information, (3) inclusion in HapMap, (4) inclusion on the Illumina HumanHap610 fixed-panel GWAS chip, and (5) novel variants discovered as a part of this effort. In total, 154 SNPs and 5 indels reported in dbSNP (http://www.ncbi.nlm.nih.gov/SNP) provide 83% bin coverage (Table 1) of the region. However, since the majority of these SNPs (N = 108) contained no frequency information within the database, this study provides further characterization for these predicted variants. A subset (N = 58) of dbSNP SNPs were characterized as a part of the International HapMap Project (http://www.hapmap.org), and only provide 50% bin coverage of the 97-kb region. An even smaller subset of SNPs (N = 16) are included on the widely used Illumina HumanHap610 fixed panel GWAS chip. These 16 SNPs only provide 28% bin coverage. Considering only the 22 common (>5% MAF) SNPs discovered by our resequencing project, these novel variants provide 33% bin coverage. All SNPs and indels taken together, regardless of source, provide 100% coverage of this region, and will be extremely useful in studies focused on this region of the genome.

Discussion

In this study, we have characterized the set of common SNP and indel variants across a 97-kb region of chromosome 10q11.2, and in the process have doubled the number of variants available for follow-up in individuals of European background. It is notable that we did not discover additional SNPs or indels in perfect LD with rs10993994, the promoter SNP most strongly associated with prostate cancer risk in two independent GWAS, which has also been shown to display a molecular phenotype consistent with decreased expression of MSMB. Consequently, our findings further point toward rs10993994 as the probable variant that at minimum partially accounts for the association with prostate cancer. Our analysis also provides a comprehensive portrait of common variation across this region. Tagging of the region with an r 2 threshold of 0.8 for variants with a MAF > 5%, 36 variants are required, but tagging at more aggressive r 2 thresholds of 0.9 and 1.0 increased the number of tags to 39 and 116, respectively.

There is increasing evidence that the microseminoprotein-β (MSMB) gene on chromosome 10q11.2 plays a role in the development of prostate cancer. GWAS studies have pointed to this region and follow-up studies have shown that altered gene expression of MSMB, as a consequence of the most significantly associated variants in the promoter, correlated with prostate cancer risk. Levels of MSMB reportedly are decreased in patients with poorer outcome and higher cancer recurrence (Nam et al. 2006; Reeves et al. 2006). Furthermore, the protein encoded by MSMB has been shown to suppress the growth and induce apoptosis of prostate cancer cells (Lokeshwar et al. 1993; Mundle and Sheth 1993) Though strong evidence suggests that alleles of rs10993994 are functionally important (Chang et al. 2009; Lou et al. 2009) and account for a portion of prostate cancer susceptibility (Chang et al. 2009; Eeles et al. 2008; Lou et al. 2009; Thomas et al. 2008), it is also plausible that other variants in this region of chromosome 10q11.2, perhaps in other genes such as NCO4, could further contribute to risk (Lou et al. 2009).

Using 454 next-generation sequencing technology (Rothberg and Leamon 2008), we have extensively characterized common genetic variation across this region of chromosome 10q11.2, which includes the MSMB gene and a large upstream block of LD that contains its proximal promoter and the SNP for which functional alleles have been described, rs10993994 (Chang et al. 2009; Lou et al. 2009). Though 24 exonic variants are reported in the dbSNP database, no polymorphic variants were confirmed within the exons of the MSMB gene. Within the ~51-kb block of LD that contains rs10993994 and the proximal promoter of MSMB, we characterized 282 genetic variants, including 94 and 49 novel SNPs and indels, respectively.

As a part of this project, we also included the sequencing of a neighboring gene, nuclear receptor coactivator 4 (NCOA4, also known as ARA70). MSMB and NCOA4 are physically located ~10 kb apart, and NCOA4 may also be of interest in prostate cancer since it is a coregulator of androgen receptor (AR) (Yeh and Chang 1996). The ARA70 protein has at least two known isoforms (α and β), both of which may have effects on prostate cancer growth though through potentially different mechanisms. ARA70α inhibits androgen-dependent cell proliferation, and ARA70β promotes proliferation and enhances prostate cancer invasion (Peng et al. 2008). In addition to observing an association with rs10993994, Chang et al. (2009) observed a potential second region of association (represented by rs10761581) in a block of LD across NCOA4 that was separated from rs10993994 by a recombination hotspot (Chang et al. 2009), though the association was not consistent across all replication studies. Common genetic variation in this region should be further explored for independent evidence of association with prostate cancer; a precedent has already been set for multiple independent associations in close proximity from prostate cancer GWAS, including 8q24 (Gudmundsson et al. 2007a; Haiman et al. 2007; Yeager et al. 2007) and HNF1B (Sun et al. 2008).

The relationship between the two contiguous genes on chromosome 10q11.2, MSMB and NCO4 could be complex. In the National Center for Biotechnology Information (NCBI) public database AceView (http://www.ncbi.nlm.nih.gov/IEB/Research/Acembly/) (Thierry-Mieg and Thierry-Mieg 2006)), also displayed within the UCSC genome browser (http://genome.ucsc.edu/) (Kent et al. 2002), MSMB and NCOA4 form a complex locus in which mRNA transcripts utilize exons from both genes. For example, expressed sequence tag (EST) DB215804.1 (Kimura et al. 2006) is comprised of exons 1 and 2 of MSMB and exons 2–4 of NCOA4. Therefore, the MSMB promoter could affect the regulation of hybrid MSMB-NCOA4 proteins. It will be important to further explore the role of genetic variation at NCOA4 and prostate cancer risk.

We have characterized common genetic variation across ~97 kb of chromosome 10 (chr10: 51,168,025–51,265,101) that is of immediate interest in prostate cancer research, since it is comprised of (1) a region of linkage disequilibrium (LD) in which a known functional polymorphism (rs10993994) resides, (2) a strong candidate gene for prostate cancer risk (MSMB), and (3) an additional gene (NCOA4) that is of interest in prostate cancer and parts of which that may, in some cases, be transcribed with exons from MSMB. Common and uncommon variants identified in this comprehensive resequence analysis of a region of chromosome 10q11.2 have established a foundation for choosing tag SNPs in follow-up studies to further map the region in an effort to nominate the most likely candidate variants for follow-up functional laboratory work to investigate the molecular basis of prostate cancer risk in this region.