|Home | About | Journals | Submit | Contact Us | Français|
Genetic variation in two members of the Toll-like receptors family, TLR4 and the gene cluster TLR6-1-10, has been implicated in prostate cancer in several studies, but the associated alleles have not been consistent across reports.
We performed a pooled analysis combining genotype data from three case-control studies, CAPS, HPFS and PLCO, with data from 3,101 prostate cancer cases and 2,523 controls. We performed imputation to obtain dense coverage of the genes and comparable genotype data for all cohorts. In total, 58 SNPs in TLR4 and 96 SNPs in TLR6-1-10 were genotyped or imputed and analyzed in the entire dataset. We performed cohort-specific analysis as well as meta-analysis and pooled analysis. We also evaluated whether the analyses differed by age or disease severity.
We observed no overall association between genetic variation at the TLR4 and TLR6-1-10 loci and risk of prostate cancer.
Common germline genetic variation in TLR4 and TLR6-1-10 does not appear to have a strong association with risk of prostate cancer.
Epidemiological, molecular and animal studies all indicate that chronic inflammation in the prostate is a risk factor for prostate cancer (1). Accordingly, genes involved in inflammation processes have been suggested to alter prostate cancer risk. Genetic variation in two members of the Toll-like receptors family, TLR4 and the gene cluster TLR6-1-10, has been implicated in prostate cancer in several association studies, but no consensus in terms of casual alleles has been reached (2–7).
We performed a pooled analysis of TLR4 and TLR6-1-10 genetic variation in a total of 3,101 prostate cancer cases and 2,523 controls from three well established studies. We also performed imputation in order to get an extensive coverage of the genes and complete data for all cohorts. In total, 58 SNPs in TLR4 and 96 SNPs in TLR6-1-10 were analyzed in all cohorts. This is the largest, most comprehensive study of the association between common genetic variation in these genes and prostate cancer risk to date.
The three participating case-control studies have been described earlier. Cancer of the Prostate in Sweden (CAPS) is a population-based case-control study of prostate cancer in Sweden collected between 2001 and 2003 (2). A total of 1,278 prostate cancer cases and 710 controls were included in this study. The Health Professionals Follow-up Study (HPFS) (3) is a case-control study nested within the ongoing Health Professionals Follow-Up Study. A total of 51,529 U.S. men aged 40 to 75 years were enrolled in 1986. For this study, 659 prostate cancer cases and 656 controls were included. The Prostate, Lung, Colon and Ovarian Cancer Screening Trial (PLCO) is a large randomized controlled trial of approximately 155,000 men and women where participants are randomized to either a screening or control arm. Enrollment lasted between 1993 and 2000. Genotype data was extracted from a genome-wide scan within the CGEMS project including 1,230 prostate cancer cases and 1,230 controls from PLCO (8). All subjects were of European ancestry. The pooled data from all three study populations consists of 3,113 cases and 2,523 controls. Of the cases, 1,297 (42%) were classified as aggressive defined as either having a Gleason Score 7–10 and/or stage C/D. A total of 1,382 (44%) of the cases and 1,100 (44%) of the controls were younger than 65 years old.
SNP selection and genotyping methods have been described earlier (2,3,8). To get complete data for all cohorts, we performed imputation using the MACH software (9), http://www.sph.umich.edu/csg/abecasis/MACH/. Imputation was performed in each cohort separately. For imputation, we used phased data from HapMap (release 21a: build 35) on individuals with European ancestry (CEU). We included SNPs spanning from 20 kb upstream of the gene (for TLR6-1-10, 20 kb upstream of TLR6) and 10 kb downstream of the gene (for TLR6-1-10, 10 kb downstream of TLR10). In total, we had imputed TLR4 genotype data for 60 SNPs in CAPS, 63 SNPs in HPFS and 60 SNPs in PLCO. For 58 of the SNPs, we had imputed genotype data from all cohorts. For TLR6-1-10, we had imputed genotype data for 97 SNPs in CAPS, 96 SNPs in HPFS and 98 SNPs in PLCO. For 96 of the SNPs, we had imputed genotype data from all cohorts.
Study-specific odds ratios and 95% confidence intervals were calculated using unconditional logistic regression. For imputed SNPs we used “dosages”, i.e. expected counts of minor alleles (fractional number between 0 and 2) as obtained from the MACH software. We adjusted all analysis for age in five years interval and cohort using indicator variables. Allele frequencies across studies were similar, indicating overall low genetic heterogeneity between populations. We used the software R for all calculations (10).
No SNP in TLR4 were associated with prostate cancer in the pooled data (Table 1) although several SNPs were nominally significant in sub-analysis of subjects younger than 65 years (P =0.02–0.05) (data not shown). For TLR6-1-10, two uncommon SNPs (MAF=3%) were nominally associated with prostate cancer risk (P=0.04, Table 2) and several uncommon SNPs (MAF<0.05) showed marginal associations with aggressive prostate cancer (P =0.05) and with late age of onset (data not shown). However, none of the associations remained significant after adjustment for multiple testing. Multi-SNP analysis using Kernel Machines (12,13) with a quadratic kernel (which allows for potential pairwise interactions) did not show any association (P =0.27 for TLR4 and P =0.23 for TLR6-1-10).
In this analysis, we used data from three large established prostate cancer case-control studies to assess association between prostate cancer and genetic variation in the TLR4 and TLR6-1-10 gene clusters. We found no overall association and furthermore, no consistent associations between TLR SNPs and prostate cancer risk between cohorts. Single study-specific associations that were observed did not remain significant after adjustment for multiple testing.
By retrospectively combining genetic marker data from already published studies we were able to increase the power to detect an association. Whilst each individual study included in this analysis had limited power, the pooled data has more than 80% power to detect a SNP with a minor allele frequency of 0.09 and a log-additive odds ratio of 1.2.
We used information from the HapMap data and imputed data from each study separately. For the SNPs that passed quality control in all three studies, we pooled the data and calculated pooled odds ratios adjusted for cohort and age. We also performed meta-analysis using both fixed and random effects and found virtually no differences in results. Even though some tests for heterogeneity were nominally significant, the consistency in minor allele frequencies across studies together with the inclusion of only Caucasians subjects, the high accuracy in genotyping across studies and the use of same reference data and parameters for imputation, suggest our pooled analyses are valid.
This study provides solid evidence against a strong association between common genetic variation in TLR4 and TLR6-1-10 in prostate cancer risk. Moreover, this study shows how imputation can be used in practice to elucidate the inconsistent findings in genetic association studies and reach a consensus about the existence or otherwise of associations.
The authors would like to thank all study participants for their generous contribution. We also thank Dr. Edward Giovannucci and Constance Chen. This work was partially funded by NIH CA098233. SL was supported by a grant from the Swedish Research Council.