Admixture Mapping of Prostate Cancer in African Americans participating in the North Carolina-Louisiana Prostate Cancer Project (PCaP)

Abstract

BACKGROUND

Few genetic risk factors have been uncovered that contribute specifically to the racial disparity in prostate cancer (CaP) observed in African Americans (AA). With the advent of Ancestry Informative Marker (AIM) single nucleotide polymorphism (SNP) panels and powerful genetic strategies such as Mapping by Admixture Linkage Disequilibrium (MALD) it is possible to discover genes that underlie ethnic variation in disease risk.

METHODS

1130 AA CaP cases enrolled in the North Carolina-Louisiana Prostate Cancer Project (PCaP) were genotyped using a 1,509 AIM SNP panel. MALD was performed using ADMIXMAP to test for linkage between CaP risk and ancestry estimates at each AIM SNP.

RESULTS

The largest increase of African ancestry was observed at marker rs12543473 (P=0.0011), located on chromosome 8q24.21, and the greatest excess of European ancestry was observed at marker rs10768140 (P=0.0004) at chromosome 11p13.

CONCLUSIONS

The study confirmed the 8q24 risk loci and identified a novel genomic region on 11p13 that is associated with CaP risk. These findings should be replicated in larger AA populations and combined with fine mapping data to further refine the novel 11p13 CaP risk loci.

INTRODUCTION

Prostate cancer (CaP) is the most common cancer among men in developed countries [1]. Despite widespread screening and improved treatment, CaP remains a major public health problem and African-American men have among the highest worldwide incidence and mortality rates [2].

CaP is a multifactorial disease with both genetic and environmental components. Quantitative estimates from twin studies indicate that 42% of CaP cases may have a heritable component [3], which is stronger than for any other type of cancer in humans [4-7]. The results from both family studies and genome-wide association studies (GWAS) indicate that many genetic loci contribute to disease risk with varying levels of penetrance [8]. The racial differences in disease prevalence suggest that some of the CaP susceptibility variants may have different risk allele frequencies in different populations (e.g., African vs. European) [9]. Recently admixed populations, such as African Americans, may have an increased probability of inheriting a chromosomal segment that contains causal variants from the ethnic population with higher disease prevalence (African). A number of CaP genetic loci have emerged from genome-wide association studies (GWAS) [10], but fewer genetic risk factors have been uncovered that contribute specifically to the racial disparity observed in African Americans (AA) [11,12].

New approaches to disease mapping in admixed populations [13] that compliment GWAS (association analyses of cases and controls within the same ancestral group) have gained attention [13,14]. The recent availability of Ancestry Informative Marker (AIM) single nucleotide polymorphism (SNP) panels (SNPs whose allele frequency vary widely by ancestral group – e.g., African, European) and new analytic tools have made possible a powerful genetic strategy known as Mapping by Admixture Linkage Disequilibrium (MALD) . MALD facilitates the discovery of genes that underlie ethnic variation in disease risk and capitalizes on the long-range linkage disequilibrium (LD) that creates large ancestral haplotype blocks, which characterize genomes of recently admixed populations, such as AAs. The average size of ancestry blocks in AA populations is much longer than the haplotype blocks of un-admixed human populations. Thus MALD is more efficient than GWAS since MALD requires 200-500-fold fewer SNPs to canvas the genome, but retains the high statistical power of association studies. Additionally control subjects are not required, because MALD compares the proportion of ancestry at each locus to the average genome-wide ancestry using affected persons only (cases). Comparisons of admixture mapping and association analysis in family, patient and population-based studies have demonstrated that each test can provide unique information and that ancestry can be more informative than direct association analysis when the causal variant has large allele frequency differences between ancestral populations [15,16].

Only two studies have reported on admixture mapping of CaP among AA [11,12]. Freedman et al. [11] in 2006 capitalized on developing AIMs panels and evaluated 1,597 CaO cases from 7 different studies, while Bock et al. [12] in 2009 evaluated 482 cases from two independent studies. While both studies confirmed the 8q24 locus, Bock et al. [12] reported several other ancestry-specific susceptibility loci. Herein we use an established AIM panel to conduct admixture analysis on a large well-characterized AA CaP cohort; AAs enrolled in the population-based North Carolina-Louisiana Prostate Cancer Study (PCaP).

SUBJECTS AND METHODS

Study population

PCaP is a population-based, case-only study designed to investigate social, individual, and tumor-level causes of racial differences in prostate cancer that has been described [17]. PCaP recruited 1130 black/ AA CaP cases between 40 and79 years of age with newly diagnosed, histologically-confirmed, adenocarcinoma of the prostate from North Carolina (NC) and Louisiana (LA). Medical records were abstracted for information related to a CaP diagnosis which included total serum prostate-specific antigen (PSA) (defined as the PSA value closest and within 1 year prior to CaP diagnosis date), tumor stage at diagnosis (stage number was derived from stage as reported in the medical record such that 1=T1a, T1c T1(not otherwise specified (NOS)), T1C; 2=T2(NOS), T2a, T2b, T2 a or b, T2c; 3=T3/4 (NOS), T3b, T3a, 3A and 4=T4) and grade (Gleason score= sum of primary and secondary Gleason grade) for aggressiveness classification. Men were classified based on clinical Gleason score, clinical stage, and PSA at diagnosis as: (1) highly aggressive (Gleason score ≥ 8, or PSA > 20 ng/ml, or Gleason score = 7, and stage cT3–cT4); (2) non-aggressive (Gleason score < 7 and stage cT1–cT2, and PSA < 10 ng/ml); or (3) intermediate aggressive (all other cases). DNA was available on 997 AA who are the subjects of this study. Age at diagnosis was derived from self-reported date of birth and date of diagnostic biopsy as indicated in the medical record and rounded to the nearest full year. Informed consent was obtained from all research subjects prior to blood and questionnaire collection and study protocols were approved by participating institution Internal Review Boards.

Genotyping

DNA was extracted from blood samples (n=811) or buccal cells (n=44) by the University of North Carolina at Chapel Hill (UNC-Chapel Hill) Biospecimen Processing Facility, or from lymphocytes immortalized by the UNC-Chapel Hill Tissue Culture Facility (n=142). Genotyping was performed by the NIH Center for Inherited Disease Research (CIDR) using the Illumina AA Admixture Panel. The SNP panel contains 1,509 ancestry-informative SNPs with averaged allele frequency difference of 0.74 between HapMap CEU and HapMap YRI population. The averaged chromosome distance between SNPs is 1,950kb. Twenty sample duplicates and 11 HapMap trios with representation of CEU and YRI ancestral groups were analyzed.

Statistical Analysis

An affected-only MALD analysis was performed in PCaP AA CaP case genotype data using ADMIXMAP software [18] to compare observed versus expected ancestry allele copies across the genome conditional on a priori ancestral allele frequencies. ADMIXMAP fits a Bayesian probability model with computationally intensive Markov chain Monte Carlo parameter estimation algorithm.

The affected-only test requires strong prior information on allele frequencies for any specified ancestry [19]. Priors, were calculated from allele frequency based on HapMap west African Yoruban (YRI) and CEPH Europeans from Utah (CEU) genotypes (http://hapmap.ncbi.nlm.nih.gov/), and ADMIXMAP software was used on genotype data obtained using the same panel of SNPs from 608 AAs who were part of a separate study to estimate ancestry-specific allele frequencies for each SNP with the HapMap allele frequency prior. The model-fit evaluation statistics - the deviance information criterion (DIC) for affected-only tests on PCaP data were 2.16×10⁶ and 2.15×10⁶, respectively for HapMap prior and AA data estimated allele frequencies prior. These results indicated AA estimated allele frequency priors produced a better model fit, and thus the estimated allele frequencies were used as priors in the final PCaP data analysis. For each research subject, the affected-only method used the score test to compare ancestry estimates at each locus to the corresponding ancestry estimates of the whole genome. The score for each locus was calculated by averaging the posterior probability of linkage with ancestry across samples. The test statistic (Z-score) was the score divided by the square root of the observed information, and thus had a standard normal distribution under the null hypothesis. A positive Z-score indicated excess African ancestry allele copies at a locus while a negative Z-score indicated excess European ancestry allele copies at a locus. Similar to Bock et al[12] and Schwartz et al[20], we reported peak regions with absolute value of Z-score greater than 3.0 (P-value ~ 0.0025).

ADMIXMAP software also can evaluate residual LD among SNPs that are not accounted for by the admixture model. SNPs with genotype data highly correlated with adjacent SNPs (score test for residual LD, P<1×10⁻⁵) were removed from final analysis. Furthermore, ADMIXMAP was used to adjust for the excess of African ancestry on the X chromosome due to historical sex asymmetry with a prior of exp(3) for the population-level odds ratio female/male in founders [21]. Once this adjustment was made, the Z scores for the X chromosome were interpreted in concert with the MALD results for autosomes. The recently built AA genetic map [22] for SNP map positions was used in the analysis to increase the precision of admixture estimates.

RESULTS

Allele calling was conducted using Illumina’s Genotyping Module version 1.0.10 in GenomeStudio 1.0.2.20706. The genotype intensity cluster plots were inspected visually for each SNP. Genotypes with an Illumina GenCall (GC) score below 0.25 were assigned as missing. Seven PCaP research subjects were excluded because of failed genotyping due to poor sample quality. In total, 990 research subjects with high quality genotype data were available for analysis, which included 447 AA from NC and 543 AA from LA. A total of 58 SNPs (3.8%) were excluded due to poor clustering pattern or parent-parent-child (P-P-C) heritability errors identified using HapMap trios. In addition, 13 SNPs were excluded by the Hardy-Weinberg equilibrium test or residual LD test P-value <1×10⁻⁵. Genotypes of 1438 SNPs for 990 AA CaP cases remained after all exclusions. The overall research subject genotyping call rate was 99.96%. The reproducibility rate was 100% based on duplicates and the overall P-P-C heritability was 99.97% based on HapMap trios. Table 1 describes the characteristics of these eligible cases by study site. The mean estimated individual proportion of African ancestry for PCaP research subjects was slightly higher in North Carolina (82%) than Louisiana (79.3%). The distributions of other clinical characteristics were similar between states.

Table 1.

Clinical characteristics of PCaP AA research subjects in data analysis

	NC	LA
No. of research subjects	447	543
Age at diagnosis
Median	60	63
Range	41-79	43-79
Mean African ancestry (%)	82.0	79.3
Median PSA ng/ml	6	6
PSA levels N (%)
< 4 ng/ml	40 (8.9)	65 (12)
4-10 ng/ml	276 (61.7)	305 (54.2)
10-20 ng/ml	74 (16.7)	83 15.3)
> 20 ng/ml	52 (11.6)	56 (10.3)
Unknown	5(1.1)	34 (6.2)
Gleason Score, N (%)
≤6	237 (53.0)	309 (56.9)
7	155 (34.7)	157 (28.9)
8-10	55 (12.3)	76 (14.0)
Unknown	0 (0)	1 (0.2)
Pathologic stage, N (%)
T1	274 (61.3)	266 (49.0)
T2	162 (36.3)	246 (45.3)
T3-T4	9 (2.0)	8 (1.5)
Unknown	2 (0.4)	23 (4.2)
Aggressiveness, N (%)
Low	203 (45.4)	244 (45.0)
Intermediate	157 (35.1)	159 (29.3)
High	87 (19.5)	110 (20.2)
Unknown	0 (0)	30 (5.5)

The distribution of Z-scores resulting from ADMIXMAP is shown in Figure 1. Details about the two SNP markers at local peaks with |Z|>3.0 (P-value ~ 0.0025) are summarized in Table 2. The most significant increase in African ancestry among AA PCaP cases are observed on chromosome 8q24.21, while the most significant increase in European ancestry are observed on chromosome 11p13 (Table 2). A more detailed examination of chromosomal regions with |Z|>3.0 is shown in Figures 2A and 2B. Z-scores for all AIM SNPs in the MALD analysis are provided in Supplemental Table 1.

Figure 1.

Genome-wide admixture mapping results in 990 AA men with CaP.

Table 2.

Summary of SNPs at regional ancestry association peaks with |Z| >3.0

Locus	Location	Position	Entire sample		Age ≤ 60		Age > 60
			Z	P	Z	P	Z	P
rs12543473	8q24.21	127850622	3.267	0.0011	2.437	0.0148	2.317	0.0205
rs10768140	11p13	35467483	−3.515	0.0004	−2.063	0.0391	−2.857	0.0043

Figure 2.

Detailed local chromosomal regions with |Z|>3.0 in PCaP MALD analysis.

Whether reported GWAS SNPs [23-33] coincided with admixture results was investigated by mapping the chromosomal position of these SNPs onto the admixture mapping results. Only 5 of the 37 GWAS regions coincided with MALD peaks of |Z|>2 (Supplemental Table 2, Supplemental Figures 1 and 2).

DISCUSSION

Results from PCaP identified a significant increase in African ancestry on chromosome 8q24.21, which confirmed two admixture mapping analyses in AA men with prostate cancer [11,12]. In the admixture mapping analysis reported by Freedman et al. in 2006 of 1,597 AA with CaP from 7 studies, a 3.8 Mb interval on the 8q24 chromosome alone emerged as a region with the greatest increased African ancestry related to CaP using ANCESTRYMAP (LOD score = 4.1 in all cases) [11,34]. Freedman et al. in 2006 [11] genotyped 15 additional SNPs among 294 additional cases to better localize the admixture peak. rs12543473 (at physical location 127,850,622 in NCBI Genome Build 35) was among these additional SNPs, which was the SNP on our AIM panel that yielded the most significant African ancestry signal. In the additional analysis by Freedman et al., this SNP displayed the second highest case-only LOD score of 3.97, second only to its neighboring SNP, rs780321 at map position 127,152,877 with LOD score 4.01. A smaller admixture mapping study by Bock et al. [12] in 2009 of 482 AA cases and 261 controls using ADMIXMAP and the Illumina AIMs panel, similar to our study, reported finding the strongest 8q24 association (though non-significant) with rs4367565 at map position131,606,887-131,607,387 (NCBI Genome Build 35) on chromosome 8 (in both a case-control and case-only analysis). When comparing our results to the more dense SNP evaluation of the 8q24 genomic region (124-132 megabases (Mb) by Freedman et al [11]. our SNP signal is located directly within the same 3.8 Mb peak at 127 Mb. In contrast the signal observed by Bock et al., determined using a much smaller sample group, is in the tail of the peak delineated by Freedmen et al at 131 Mb with low probability density [11,12]. Our results present a somewhat bimodal distribution at 8q24, which likely reflects lack of refinement (6-7 cM) of the MALD map rather than a significant observation. In addition to excess African ancestry on 8q24, Bock et al. reported a region on chromosome 5q35 (rs7729084 and rs12474977, case-only analysis). As indicated by Bock et al., this region on chromosome 5 also demonstrated higher African ancestry in our analysis (rs7729084, Z score=1.29), though the finding did not reach statistical significance.

Several regions of the genome with European ancestry have been found to increase CaP risk in admixture mapping of AAs. Bock et al. reported an association with 7q31 (rs2141360), and although a higher proportion of European ancestry was found at this locus, the association did not reach statistical significance (rs2141360 Z-score = −1.23). The most significant increase in European ancestry was found at a novel locus on chromosome 11p13 with Z score well below −3 (rs10768140); this region has not been reported in admixture mapping studies of AA CaP cases to-date [11,12]. This SNP is located within the intron of peptidase domain-containing protein associated with muscle regeneration 1 (PAMR1) that may play a role in regeneration of skeletal muscle and is down-regulated in muscle cell lines from patients with Duchenne muscular dystrophy (http://www.ncbi.nlm.nih.gov/). Given the function of this gene, the AIM marker probably represents a marker for an associated CaP locus within the ancestral region rather than a functional variant. No small nuclear or microRNAs are reported in a region (1 Mb) surrounding this SNP (UCSC Genome Browser Build 37 [35-41]. Several other confirmed genes lie within a 1 Mb region surrounding rs10768140 (35 MB – 36 MB on chromosome 11). CD44, a cell-surface glycoprotein involved in cell-cell interaction including cell migration is located telomeric to PAMR1 (http://genome.ucsc.edu/). CD44 interacts with ligands, such as matrix metalloproteinases, and participates in tumor metastasis (UCSC Genome Browser) in cancers including prostate [42]. Additionally, CD44 is a stem cell marker for breast and CaP [43,44]. Similar to the fine mapping study by Freedman et al. that refined an 8 Mb region to a peak of 3.8 Mb [11], further fine mapping in this region is needed to identify the specific signal location. While genes within a 1 Mb region surrounding our AIM signal were reported on, genes in more distant locations could be relevant, especially because the average ancestry block size observed in African Americans is 17-20 centimorgans (cM) [34,45-49] or approximately 17-20 Mb on the physical genetic map. Bock et al. did not report excess European ancestry on chromosome 11, however it is unclear if chromosome 11 was evaluated since the genome-wide admixture plot did not delineate several autosomes including chromosomes 11, 15, 18 and 20 and supporting documentation was not provided detailing the Z scores for all genotyped SNPs. Even though the same Illumina AIM panel was used by Bock et al., there may have been a lack of coverage across specific genomic regions including chromosome 11 after genotypes were removed that had poor call rates (N=188).

Random associations may occur within MALD and may be dominated by the signal from a single SNP, producing a narrow peak that is often near the tail of a chromosome. The 11p13 peak was relatively wide and did not include SNPs from the tail of the chromosome; this peak retained a |Z| score greater than 3 even after removal of the SNP with the strongest association (|Z|=3.26 2 for adjacent SNP rs2553779 and |Z|=3.255 for adjacent SNP rs6484807).

Both 8q24, demonstrating excess African ancestry, and 11p13, demonstrating excess European ancestry, may harbor CaP susceptibility alleles that contribute to the increased CaP risk of AAs. The effect size of the 8q24 loci is large and should be detected in both GWAS and ancestry analysis and harbor alleles that contribute to both ancestry-specific and general CaP case association. The lack of identification of the 11p13 region among European GWAS analyses may be explained on several levels. First, the effect size may be less than that of 8q24 and the allele frequency differences much larger, which makes this locus easier to identify by ancestry mapping and more difficult by GWAS because of the penalties incurred in GWAS for multiple tests [14]. In addition, GWAS is not robust to allelic heterogeneity, a likely feature of CaP, while admixture mapping is. If different risk variants at the same locus within the 11p13 region are responsible for CaP risk, this region may be more readily detected by MALD.

A small but well-established AIM panel was used to infer African and European ancestral segments. New analytic methods that leverage both MALD and GWAS in either joint analyses or 2-stage design are emerging [14,50-53]. These newer designs will capitalize on the efficiency of MALD by reducing statistical penalties in the discovery phase, while providing refinement of the genetic susceptibility loci by direct fine mapping conducted in the second phase.

In conclusion, this study confirms the 8q24 risk loci and identifies a novel genomic region on 11p13 that is associated with CaP risk in AAs. These MALD findings should be replicated in larger AA populations and combined with fine mapping data to further refine this novel CaP risk loci.