Fine-mapping of IL16 gene and prostate cancer risk in African Americans

Abstract

Background

Prostate cancer (Pca) is the most common type of cancer among men in the United States, and its incidence and mortality rates are disproportionate among ethnic groups. While genome-wide association studies of European descents have identified candidate loci associated with Pca risk, including a variant in IL16, replication studies in African Americans (AAs) have been inconsistent. Here we explore SNP variation in IL16 in AAs and test for association with Pca.

Methods

Association tests were performed for 2,257 genotyped and imputed SNPs spanning IL16 in 605 AA Pca cases and controls from Washington, DC. Eleven of them were also genotyped in a replication population of 1,093 AAs from Chicago. We tested for allelic association adjusting for age, global and local West African ancestry.

Results

Analyses of genotyped and imputed SNPs revealed that a cluster of IL16 SNPs were significantly associated with Pca risk. The strongest association was found at rs7175701 (P=9.8 × 10⁻⁸). In the Chicago population, another SNP (rs11556218) was associated with Pca risk (P=0.01). In the pooled analysis, we identified three independent loci within IL16 that were associated with Pca risk. SNP eQTL analyses revealed that rs7175701 is predicted to influence the expression of IL16 and other cancer related genes.

Conclusion

Our study provides evidence that IL16 polymorphisms play a role in Pca susceptibility among AAs.

Impact

Our findings are significant given that there has been limited focus on the role of IL16 genetic polymorphisms on Pca risk in AAs.

Introduction

Prostate cancer (Pca) is the most common type of cancer among men in the United States, and its disproportionately higher rate of incidence and mortality among African Americans (AAs) are very likely due to a combination of genes and environmental factors. While genome-wide association studies (GWAS) in men of European descents have identified candidate loci associated with high Pca risk (1-6), the same locus tend not to be replicated in studies of AAs (7, 8). Moreover, GWAS have not yet produced many insights into Pca disparities (9-11). For instance, a single nucleotide polymorphism (SNP) in IL16, rs4072111, was associated with Pca aggressiveness in European Americans (EAs) (1), but a subsequent study in AAs did not replicate the result (7). Given that currently available GWAS panels vary in genome coverage across ethnic groups due to marker selection and linkage disequilibrium (LD) differences, the ‘exact’ replication of index SNPs identified in people of European-descent is often problematic in populations with lower LD (12). In populations with lower LD such as AAs, fine-mapping of SNPs by direct genotyping or by imputation combined with deep re-sequencing provides for high resolution assessment of variation and is useful for ‘local’ replication (replicating associations between phenotype and variants in the same gene/locus as previously reported index SNPs), and finding causal markers (13, 14).

Recent research has demonstrated the link between inflammatory processes and the development of cancer cells. Proinflammatory mediators, such as cytokines, chemokines, prostaglandins, and nitric oxide along with enzymes that respond to these factors are involved in cancer cell proliferation and progression (15, 16). It also has been hypothesized that inflammatory processes can cause oxidative damage in prostate epithelial cell that can promote proliferative inflammatory atrophy, a precursor of prostatic intraepithelial neoplasia and Pca (17-19). Cytokines are key regulator of inflammation, but over-expression of cytokines during chronic inflammation creates a microenvironment that facilitates initiation and progression of tumor cells (16, 20).

Both immune cells and tumor cells produce cytokines, including IL-16, a chemoattractant for immune cells with CD4 co-receptors, including T cells, monocytes, macrophages, eosinophils, and dendritic cells. Signaling of IL-16 is mediated by the CD4 co-receptor, and IL-16 activates these CD4 immune cells. As in many other organs, T cells, B lymphocytes, macrophages, and mast cells are found in prostate, and greater numbers of CD4 T cells than CD8 cells are found in inflamed benign prostatic hyperplasia (17, 21). Although specific functions of IL-16 in immune response are not well understood, IL-16 is considered as a key mediator of inflammatory response (22) and activates tumor necrosis factor-α (TNF-α), IL-1β, IL-6, and IL-15 (23). These cytokines are known to promote tumor cell growth (15).

Here we explore SNP variation in IL16 in AAs and test for association with Pca. Lower LD across the gene region in African-descent populations may explain the lack of replication of the IL16 SNP in AAs. Thus, to search for associated SNPs in AAs, first, we tested association of genotyped and imputed SNPs in and around IL16 in AA Pca cases and controls from Washington, D.C. Second, we selected 11 SNPs for more close examination and attempted to replicate the results using an independent group of AA Pca cases and controls from Chicago, IL.

Materials and Methods

Subjects

The discovery sample set consists of 260 Pca cases and 345 controls from Washington, D.C., which were a part of a recent GWAS for Pca in AAs (24). These participants were recruited through the Division of Urology at Howard University Hospital in Washington, D.C. (25, 26). The replication sample set includes 543 Pca cases and 550 controls recruited from three Chicago-area hospitals (27). All the participants are unrelated self-identified AAs (Table 1).

Table 1.

Subject characteristics.

	Cases	Controls	P 1
Discovery – Washington, D.C.
Number of Subjects	260	345
Age (mean and SD)2	64.3 (9.7)	58.7 (10.5)	<0.001
PSA in ng/mL (mean and SD)2	48.9 (140.9)	2.6 (6.9)
Global WAA (mean and SD)2	81.2 (15.8)	79.5 (18.0)	0.26
Local 15q26.3 WAA (mean and SD)	80.6 (18.1)	79.6 (18.2)	0.51
Replication – Chicago, IL
Number of subjects	543	550
Age (mean and SD)2	63.5 (7.3)	58.2 (10.4)	<0.001
PSA in ng/mL (mean and SD)2	45.0 (195.4)	3.3 (4.9)
Global WAA (mean and SD)2	79.1 (15.0)	78.7 (14.7)	0.81
Pooled
Number of subjects	803	895
Age (mean and SD)	64.0 (8.9)	58.5 (10.4)	<0.001
PSA in ng/mL (mean and SD)	47.7 (159.1)	2.9 (6.2)
Global WAA (mean and SD)	80.3 (15.5)	79.1 (16.5)	0.27

¹T-test P values

²Mean age, Prostate-specific antigen (PSA) levels, and % West African ancestry (WAA) is not statistically different between discovery and replication cases (P>0.05).

Genotyping

Subjects in the discovery sample set from Washington, D. C. were genotyped as a part of a large GWAS consortium for Pca in AAs using the Illumina Infinium 1M-Duo bead array (24). We ranked the top 10% of SNPs (out of a total of 1,047,986 SNPs) associated with Pca, and the IL16 variant, rs7175701, was one of these top ranking SNPs. From the GWAS data, 264 SNPs in IL16 and 500 Kb region (250 Kb up and downstream) around IL16 (81239219-81855109) were available for analysis.

We then selected 11 SNPs from the GWAS panel and genotyped them in the replication sample set. Seven of these 11 SNPs were chosen due to very low P-values in the discovery data set, including rs7175701 and six SNPs in high LD with rs7175701. Four of the seven SNPs are predicted to affect IL16 expression (28). In addition, we included rs4072111, rs11556218, rs11325, and rs1131445. rs4072111 was identified in the previous GWAS among EAs (1); rs11556218 is associated with colorectal and gastric cancer and nasopharyngeal carcinoma risk among Chinese (29, 30); and rs11325 and rs1131445 are in the 3′UTR and are predicted to affect miRNA binding (31). Genotyping of the replication samples was performed using iPLEX Sequenom MassARRAY. All the SNPs used in the analyses had genotyping success rates of > 95%.

In addition, we estimated of global West African ancestry (WAA) using a previously verified panel of 105 unlinked ancestry informative markers (AIMs) in both sample sets (32). To estimate local 15q26.3 WAA, we used HapMap3 YRI and CEU data and selected 42 AIMs within 15q26.3, which were genotyped in the Washington, D.C. samples. These 15q26.3 AIMs show large frequency differences between the putative ancestral populations of AAs (YRI and CEU) (δ >0.7 and F_ST >0.5), and were weakly linked or completely unlinked in our Washington, D.C. samples (r² <0.1).

Imputation

Variants within IL16 and 2 Mb region (1 Mb up and down stream) around IL16 were imputed in our discovery sample set using IMPUTE2 and 1000 Genomes Phase I (interim) data as reference panels (33). Instead of choosing reference panels which are proxies for putative ancestral populations, IMPUTE2 uses multi-population reference panels and looks for shared haplotypes between reference panels and study samples regardless of their ancestry (34). In this region, 840 SNPs were genotyped in the GWAS, but after imputation, there were 32,613 variants. For the association analyses, SNPs within the IL16 gene and 500 Kb region (250 Kb up and down stream) around IL16 were included after removing low-quality, poorly imputed variants (information metric < 0.5, variants with more than 10% missing genotypes, HWE P≤0.001) and rare variants (MAF < 0.01). Low-frequency variants (MAF between 0.05-0.01) were included in the analyses to assess whether these low-frequency variants which are rare or monomorphic in European populations and were not included in the Illumina genotyping array show strong association in AAs. After QC filtering, a total of 2,257 (1,992 imputed and 264 genotyped) SNPs were included for association tests. In order to validate the imputation results, first, we masked 23 genotyped SNPs, including rs7175701, performed imputation, and then tested for association. Secondly, we performed a cross-validation analysis using MACH and HapMap data (35, 36) (Supplementary Methods).

Statistical analyses

For QC, departure from Hardy-Weinberg Equilibrium (HWE) was tested for all the genotyped SNPs in both discovery and replication sample sets. None of the genotyped SNPs showed deviation from HWE at P < 0.001. We tested for allelic association and calculated odds ratios and 95% confidence intervals (CI) using logistic regression, adjusting for age at the time of diagnosis and global WAA in discovery, replication, and pooled data sets. For the analyses of the discovery sample set, we performed additional association tests adjusting for local WAA in addition to age and global WAA. Empiric P-values were obtained using adaptive permutation procedures implemented in PLINK. Heterogeneity of disease and SNP associations between the discovery and replication sample sets was tested using the χ² and Breslow-Day test. In the pooled data set, additional statistical tests were performed by stratifying the subjects based on their age and PSA levels at the time of diagnosis. For tests of association stratifying by age, we used the following schemes; 1) all cases and controls with age ≥ 60; 2) cases with age ≥ 60 and all controls; and 3) cases with age < 60 and all controls. For the tests of association stratifying by PSA levels, we also used three schemes; 1) cases with PSA ≤ 10 ng/mL and all controls; 2) cases with PSA > 10 ng/mL and all controls; and 3) cases with PSA ≤ 10 ng/mL and cases with PSA > 10 ng/mL. The tests for association were performed using PLINK (37).

Both local 15q26.3 and global WAA were estimated separately from the AIMs genotype data using the Bayesian Markov Chain-Monte Carlo (MCMC) method implemented in the program STRUCTURE 2.1 (38, 39). STRUCTURE 2.1 was run under the admixture model using prior population information and independent allele frequencies. We ran the MCMC method using K=2 parental populations and a burn-in length of 30,000 for 70,000 repetitions.

LD patterns were examined using Haploview (40). A LD plot of the top-ranked 30 imputed and genotyped SNPs that were strongly associated with Pca was made using 1000 Genomes Phase 1 YRI data. The correlations between each of 11 selected SNPs were also calculated and visualized on LD plots. LD plots of the discovery and replication controls were compared to LD plots of YRI and European populations. 1000 Genomes Phase 1 genotype data of unrelated YRI and European (CEU, FIN, GBR, and TSI) individuals were obtained from the ENGINES SPSmart SNP database (41). Genotype data from the four European populations were combined. Haplotype blocks were constructed using the Gabriel et al. method (42).

Results

Mean genetic ancestry estimates were similar for both local WAA on 15q26.3 (mean 80.1%) and global WAA (80.2%; Kendall’s τ=0.444; P<0.001) for our discovery sample set. There was no statistical difference in local WAA between cases and control.

The discovery sample set was a part of the Pca GWAS in AAs (24), and the IL16 variant, rs7175701, was one of the top-ranking SNPs (P=5.7 × 10⁻⁷) in this sub-group of AAs (Supplementary Table 1). Our association tests of the 264 SNPs in the IL16 region in the discovery sample set revealed a cluster of 10 SNPs within the gene that were associated with Pca risk (Supplementary Figure 1). These SNPs remained significant even after permutation testing, with rs7175701 showing the strongest association (P_EMP =1.0 × 10⁻⁶).

Then, we closely examined the association of the 11 selected SNPs genotyped in the GWAS. SNP rs7175701 and six SNPs around it showed significant association (Table 2). When association tests were performed adjusting for local 15q26.3 WAA in addition to age and global WAA, the overall association was stronger. rs7175701 exhibited the strongest association and the P-value was lower (P=9.8 × 10⁻⁸). After adjustment, the second most significantly associated genotyped SNP, rs8026245, also had slightly lower P-value (P=2.0 × 10⁻⁵). When a multivariate analysis was conducted to test the association conditioning on rs7175701, rs4616256 remained significant (P=0.04), suggesting that multiple independent risk causing alleles are segregating at IL16. This association remained significant (P=0.04), even after adjusting for local WAA. On the other hand, rs7175701 and four SNPs around it were significant, after conditioning on rs4616256. These four SNPs are in strong LD with rs7175701, while rs4616256 is weakly linked to rs7175701 (r²=0.07) (Supplementary Fig. 2a).

Table 2.

Results of association tests for Washington, D.C. discovery samples.

			OR	Asymptotic	Empirical	Adjusting for local WAA	Multivariate Analysis6
SNP	Position1	MA2	(95% CI)	P-values3	P-values4	P-values5	rs7175701	rs4616256
rs12907134	81525319	A	0.66 (0.50-0.87)	3.9×10−3	4.9×10−3	1.6×10−3	0.78	0.02
rs7179134	81533378	A	1.62 (1.23-2.13)	5.3×10−4	8.1×10−4	3.2×10−4	0.43	1.4×10−3
rs7180245	81535149	A	1.66 (1.26-2.18)	2.9×10−4	3.9×10−4	1.8×10−4	0.31	8.1×10−4
rs7175701	81558623	C	1.98 (1.51-2.58)	5.7×10−7	1.0×10−6	9.8×10−8	NA	1.8×10−5
rs8026245	81561001	G	1.73 (1.33-2.24)	3.7×10−5	2.9×10−5	2.0×10−5	0.96	7.0×10−5
rs11637363	81567973	C	1.55 (1.18-2.03)	1.4×10−3	1.6×10−3	6.3×10−4	0.21	0.38
rs4616256	81575754	A	1.85 (1.29-2.65)	7.8×10−4	1.1×10−3	6.3×10−4	0.04	NA
rs4072111	81578139	A	0.70 (0.39-1.25)	0.23	0.24	0.21	0.28	0.38
rs11556218	81598269	G	1.27 (0.93-1.74)	0.14	0.16	0.11	0.12	0.23
rs11325	81601340	T	1.09 (0.83-1.42)	0.53	0.67	0.36	0.63	0.90
rs1131445	81601782	C	0.85 (0.65-1.11)	0.22	0.26	0.24	0.39	0.41

¹Base pair postion on the chromosomal is based on NCBI build 37/hg19.

²Minor alleles.

³Asymptotic P-values were estimated using logistic regression, adjusting for age and global WAA.

⁴Empiric P-values were obtained by adaptive permutation, adjusting for age and global WAA.

⁵Asymptotic P-values were estimated using logistic regression adjusting for age, local 15q26.3 WAA, and global WAA.

⁶P-values of logistic regression adjusting for the SNPs, age, and global WAA.

In the combined set of imputed and genotyped SNPs (2,257), we found additional SNPs that show strong association with Pca, but rs7175071 still had the lowest P-value (Fig. 1 and Supplementary Table 2). The three top-ranked SNPs are adjacent to each other, strongly linked (r²>0.977), and located in the same haplotype block (Supplementary Figure 3). Many other SNPs are also strongly or moderately in LD with rs7175071. After conditioning on rs7175071 in logistic regression analysis, we found that many variants were not significant or considerably less significant, except for two (rs16972963 and rs12899457 with P-values of 6.8 × 10⁻⁶ and 8.0 × 10⁻⁴ respectively). These two variants are localized in a gene desert area over 200 Kb downstream of IL16. The association tests of imputed and genotyped SNPs adjusting for local WAA in addition to age and global WAA ancestry remained very similar, and the top three most significantly associated SNPs were the same. Moreover, we performed a cross-validation experiment using MACH and HapMap YRI and CEU data as reference panels, and the four top-ranked SNPs were exactly the same (Supplementary Note and Supplementary Figure 4).

Figure 1.

Log₁₀ P-value plot of genotyped and imputed SNPs in IL-16 (±250kb) region in Washington, D.C. AA discovery samples. The position of the IL-16 is indicated by two vertical lines.

Results of the analysis of the replication sample set for the 11 selected SNPs contrasted with the results of analysis of the discovery sample set (Table 3). None of the SNPs that showed strong association with Pca risk in the discovery sample set were significantly associated with Pca risk in the replication sample set. Instead, one missense mutation near the end of IL16 gene (rs11556218), which was not associated with Pca in the discovery sample set, was significantly associated with Pca risk (P=0.01).

Table 3.

Results of association tests for Chicago replication samples.

			OR	Asymptotic	Empirical
SNP	Position1	MA2	(95% CI)	P-value3	P-value4
rs12907134	81525319	A	0.13 (0.68-1.23)	0.55	0.75
rs7179134	81533378	A	1.01 (0.74-1.39)	0.93	0.86
rs7180245	81535149	A	0.96 (0.70-1.33)	0.82	0.75
rs7175701	81558623	C	1.03 (0.78-1.38)	0.82	0.86
rs8026245	81561001	G	0.91 (0.67-1.24)	0.55	0.86
rs11637363	81567973	C	1.09 (0.82-1.46)	0.55	0.86
rs4616256	81575754	A	1.18 (0.79-1.75)	0.42	0.75
rs4072111	81578139	A	0.90 (0.49-1.66)	0.74	0.86
rs11556218	81598269	G	1.63 (1.12-2.38)	0.01	0.008
rs11325	81601340	T	1.20 (0.89-1.62)	0.22	0.25
rs1131445	81601782	C	0.84 (0.61-1.15)	0.28	0.28

¹Base pair postion on the chromosomal is based on NCBI build 37/hg19.

²Minor alleles.

³Asymptotic P-values were estimated using logistic regression, adjusting for age and global WAA.

⁴Empiric P-values were obtained by adaptive permutation, adjusting for age and global WAA.

In the pooled analysis of the 11 SNPs, the seven SNPs that were strongly associated in the discovery sample set and the SNP that was associated with Pca risk in the replication sample set remained significant (Table 4). rs7175701 again was the most significant with an odds ratio of 1.48 (95% CI 1.22-1.79). To test if these SNPs are independently associated with Pca, we tested for association conditioning on rs7175701, and we found that two SNPs, rs4616256 and rs11556218, remained significant. When the association was tested conditioning on rs11556218, seven other SNPs were significant. When we conditioned on both rs7175701 and rs11556218, only rs4616256 remained significant. These three SNPs were weakly correlated with each other (r² < 0.12) in our discovery and replication samples.

Table 4.

Results of association tests in the pooled analysis.

							Multivariate Analysis5
SNP	Position1	MA2	OR (95% CI)	Asymptotic P-value3	Empirical P-value4	rs7175701	rs11556218	rs7175701 and rs11556218
rs12907134	81525319	A	0.76 (0.62-0.94)	0.01	0.01	0.99	0.02	0.95
rs7179134	81533378	A	1.33 (1.09-1.63)	0.006	0.005	0.62	0.007	0.61
rs7180245	81535149	A	1.33 (1.08-1.63)	0.007	0.008	0.56	0.006	0.52
rs7175701	81558623	C	1.48 (1.22-1.79)	7.52×10−5	5.47×10−5	NA	1.23×10−4	NA
rs8026245	81561001	G	1.33 (1.09-1.61)	0.004	0.005	0.73	0.002	0.95
rs11637363	81567973	C	1.32 (1.08-1.60)	0.005	0.004	0.39	0.03	0.51
rs4616256	81575754	A	1.50 (1.16-1.96)	0.002	0.002	0.04	0.001	0.01
rs4072111	81578139	A	0.79 (0.52-1.20)	0.26	0.21	0.31	0.21	0.20
rs11556218	81598269	G	1.42 (1.11-1.80)	0.005	0.005	0.02	NA	NA
rs11325	81601340	T	1.14 (0.94-1.39)	0.20	0.25	0.78	0.85	0.25
rs1131445	81601782	C	0.84 (0.69-1.03)	0.10	0.17	0.18	0.49	0.650

¹Base pair postion on the chromosomal is based on NCBI build 37/hg19.

²Minor alleles.

³Asymptotic P-values were estimated using logistic regression, adjusting for age and global WAA.

⁴Empirical P-values were obtained by adaptive permutation, adjusting for age and global WAA.

⁵P-values of logistic regression adjusting for the SNPs, age, and global WAA.

Since the association tests in discovery and replication sample sets showed striking differences, we examined minor allele frequency (MAF) differences between cases and controls in both sample sets and tested for heterogeneity between the two sample sets (Table 5). In the discovery samples, the MAF of rs7175701 and the SNPs around it are very different between the cases and controls, but the MAF of these SNPs are similar between cases and control in the replication sample set. The MAF difference between cases and controls for rs11556218 is larger in the replication sample set, but consistent in the two sample sets. Table 5 reveals the results of the test of heterogeneity. Heterogeneity in SNP association was observed for five of the SNPs with P_H < 0.10.

Table 5.

Minor allele frequency (MAF) in cases and controls and results of the test for heterogeneity.

			Discovery		Replication		Pooled		Heterogeneity
SNP	Position1	MA	Cases	Controls	Cases	Controls	Cases	Controls	P-value2
rs12907134	81525319	A	0.29	0.38	0.34	0.38	0.31	0.38	0.19
rs7179134	81533378	A	0.43	0.34	0.36	0.36	0.41	0.35	0.03
rs7180245	81535149	A	0.43	0.33	0.35	0.36	0.40	0.34	0.01
rs7175701	81558623	C	0.59	0.43	0.51	0.49	0.56	0.46	0.002
rs8026245	81561001	G	0.46	0.35	0.37	0.37	0.42	0.36	0.005
rs11637363	81567973	C	0.48	0.39	0.44	0.42	0.46	0.40	0.11
rs4616256	81575754	A	0.21	0.13	0.17	0.16	0.19	0.14	0.04
rs4072111	81578139	A	0.05	0.06	0.06	0.06	0.05	0.06	0.50
rs11556218	81598269	G	0.23	0.19	0.22	0.16	0.22	0.18	0.35
rs11325	81601340	T	0.43	0.41	0.43	0.38	0.43	0.40	0.64
rs1131445	81601782	C	0.28	0.31	0.29	0.32	0.28	0.31	0.96

¹Base pair postion on the chromosomal is based on NCBI build 37/hg19.

²Test of heterogeneity between the discovery and replication sample set based on χ² statistics. Very similar results were obtained from the Breslow-Day tests

We performed additional association tests stratifying the subjects by age and PSA levels in the pooled sample set to control the confounding effects of age and PSA levels. Although the association was not as strong due to smaller sample size, the results tended to show a similar pattern of association, and rs7175701 showed the strongest association in three of these stratified analyses (P=6.7 × 10⁻⁵, for all cases and controls with age ≥ 60; P=1.4 × 10⁻⁵, for cases with age ≥ 60 and all controls; P=2.8 × 10⁻⁵ for cases with PSA ≤ 10 ng/mL and all controls). An interesting pattern was observed for SNP association analyses of cases with PSA > 10 ng/mL and all controls, two SNPs were significantly associated with Pca risk (P=0.02 for rs4616256 and P=0.03 for rs1131445). Moreover, when cases (PSA ≤ 10 ng/mL) and cases (PSA > 10 ng/mL) were analyzed, only rs12907134 was significant (P=0.01). No SNP was significantly associated with Pca risk when the analysis was restricted to younger Pca cases (age < 60) and all controls.

Discussion

Several GWAS have found candidate loci for Pca, but the majority of these studies were conducted in men of European descent. Only one GWAS has been conducted to find Pca risk loci in men of African descent (24). With the exceptions of the strong association of 8q24 and 17q12 variants, association studies in men of African ancestry have not replicated the SNPs identified in men of European ancestry (7, 8, 43-45). The possible explanations for this failure include differences in LD patterns, small sample size, heterogeneous clinical characteristics, and environmental differences. Fine-mapping of previously identified regions has found other SNPs that show stronger association with Pca than the GWAS SNPs (46). Thus, this study attempted to replicate the GWAS finding by fine-mapping and using larger AA samples than many other replication studies.

SNP rs4072111 identified first in the GWAS in EAs (1) was not significantly associated with Pca risk in either of our AA sample sets. Only two replication studies, one performed in AAs and the other in EAs, examined the association of this SNP with Pca, but neither study found significant association (7, 47). Differences in the LD pattern between African- and European-descent populations probably explain these different results (Supplementary Fig. 2a-2d). In the 1000 Genomes data for European populations, rs4072111 is not in high LD with other IL16 SNPs that we examined, but in general these SNPs are in higher LD in European populations than in African-descent populations.

Although the ‘exact’ replication of rs4072111 was not observed, we were successful in ‘local’ replication, revealing that several other IL16 variants are independently associated with Pca risk in two independent samples of AAs. In the discovery sample set, the association tests of genotyped and imputed SNPs revealed that rs7175701 (intron 4) had the strongest signal of association. The genotyped SNP, rs4616256 (intron 9), was also significantly associated, even after conditioning on rs7175701. The association of these two intronic SNPs with cancer have not been reported previously, but they are associated with IL16 gene expression in an expression quantitative trait loci (eQTL) GWAS (28). Although association tests of imputed and genotyped SNPs revealed that no imputed SNPs were more strongly associated with Pca than rs7175701, we did observe several imputed SNPs which remained significant even after conditioning on rs7175701, including two SNPs (rs16972963 and rs12899457) in a gene desert region over 200 Kb downstream of IL16. When we attempted to replicate the findings with carefully chosen SNPs, we found evidence of another independent risk-causing SNP for Pca, rs11556218. The rs11556218 T/G polymorphism is located in exon 6 and is a missense substitution between asparagine (Asn) and lysine (Lys). In Chinese population, the G allele of rs11556218 is associated with increased risk of colorectal and gastric cancer (29), and nasopharyngeal carcinoma (30). In the pooled sample set, three weakly linked SNPs (rs7175701, rs4616256, and rs1155218) were independently associated with Pca.

The contrasting results of the association tests between our two clinical sample sets probably resulted from a combination of factors and illustrate limitations of this study. First, sample size for both clinical sample sets was small and there may not have been sufficient statistical power to show consistent association. A follow-up study that includes more subjects would increase statistical power.

Secondly, the differences in association could be attributed to differences in WAA proportion in two sample sets. The C allele of rs7175701 is more common in West African (0.57 in HapMap YRI) than non-African populations (0.46 in ASW and 0.24 in CEU), and more Washington, D.C. Pca cases had high global WAA estimates (> 90%) than Chicago Pca cases (36.5% in Washington, D.C. and 23.6% in Chicago cases) (Supplementary Fig. 5). The frequency of rs7175701 allele C, however, was not different between the discovery and replication sample sets (50% in discovery cases and controls combined and 49% in the replication set), and the mean global WAA between discovery and replication cases was not statistically different (P=0.19). The heterogeneity in global WAA proportion in the two data sets could be explained by the enrichment of local 15q26.3 WAA in Washington, D.C. samples, so we performed association tests further adjusting for local WAA in addition to global WAA. When local WAA was adjusted, the association of rs7175701 became stronger (P=9.8 × 10⁻⁸, compared to P=5.7×10⁻⁷). While we controlled for global WAA in our analyses, we also explored the association using only subjects with global WAA greater than 80% in the pooled data set. The SNP rs7175701 was still significantly associated with Pca risk (P=0.002), again suggesting that the association between rs7175701 and Pca risk was not confounded by population stratification.

Thirdly, there are multiple pathways to Pca initiation and progression, so the differences in the clinical characteristics between two sample sets could explain the differing results. The Chicago replication samples were recruited more recently than the Washington, D.C. discovery samples, and Pca was detected through more aggressive PSA screening with a lower cutoff for biopsy in some of replication samples. The characteristics of cases in the discovery and replication sample sets are, however, similar and the mean age and PSA levels between two sample sets are not statistically different (P=0.336 for age and P=0.342 for PSA). We also noticed that the replication controls had slightly higher mean PSA, so we tested association excluding control subjects with PSA > 2.5 ng/mL. The results of association tests were the same as the results of tests including all the replication controls and rs11556218 showed the strongest association. The stratified analysis in pooled dataset based on PSA levels showed similar results with the strongest association of rs7175701, confirming the validity of our results despite the heterogeneity between the sample sets.

To further understand the role of rs7175701 in gene expression and interaction, we browsed SCAN, a database of genetic and genomic data on eQTL (48), because the true causal loci are likely to be eQTLs (49). SCAN predicts that this SNP affects the expression of IL16 and eight other genes in HapMap CEU and a total of 18 genes in YRI lymphoblastoid cell lines. Five of the 18 genes (DNMT3A, ITGA1, SLC30A1, SSTR2, and TRIM35) are expressed in Pca cells and another gene ZNF281 could be a target of SOX4, a transcription factor that is over-expressed in tumor cells (50). Seven other genes (FKRP, HCP5, PELO, PPARGC1B, PRCC, RHBDD2, and ZMAT3) are expressed or involved in other types of cancer. It is noteworthy that zinc is involved in the function of three of these genes, SLC30A1 (or ZNT1, Zinc transporter), ZMAT3 (Zinc finger protein), and ZNF281 (Zinc finger protein). Zinc may play roles in Pca development (51, 52) and zinc transporters are down-regulated in Pca cells of AAs (53). Interestingly, only one gene (ITGA1) has a possible role in inflammatory response. When we searched the SCAN database for rs4616256 and rs11556218, SCAN did not predict that these two SNPs affect expression of any gene that is expressed in Pca cells or other types of cancer cells.

Effects of IL16 variants on expression of IL16 in Pca cells and other tumor cells and the association of IL16 variants with Pca aggressiveness and clinical characteristics deserve further investigation. The expression of IL16 in prostatectomy tumor specimens from Pca patients is positively correlated with Gleason score and pathological stage (54), and high expression of IL16 has been observed in other types of tumor cells (55, 56). The serum levels of IL-16 protein are also elevated in patients with other types of cancers, especially patients in advanced stages of cancer (57, 58).

Although the exact mechanism is unclear and additional functional studies are warranted, our study provides evidence that IL16 polymorphisms play a role in Pca susceptibility in AAs. Many of the significantly associated SNPs are linked to rs7175701, which is an eQTL that is associated with expression of genes related to Pca or other types of cancers. Another associated variant, rs11556218, consists of an amino acid change that may affect function of the gene. Our findings are significant given that there has been limited focus on the role of IL16 genetic polymorphisms on Pca risk. To our knowledge, this is the first study that demonstrated the association of these variants with cancer in African-descent populations. Identifying genetic variants affecting the expression and interaction of IL16 and other gene expressed in tumor cells is important for understanding the role of IL16 in tumor growth.