Association between Toll-Like Receptor Gene Cluster (TLR6, TLR1, and TLR10) and Prostate Cancer

Background: Chronic inflammation may be a risk factor for prostate cancer. Previously, we found significant associations between single nucleotide polymorphisms (SNPs) and haplotypes in Toll-like receptor (TLR) 4 and the risk of prostate cancer. TLR6, TLR1, and TLR10 are also involved in the pathogen-mediated inflammation pathway. A Swedish study observed associations between sequence variants in the TLR6-TLR1-TLR10 gene cluster and the risk of prostate cancer. We assessed if genetic polymorphisms of this gene cluster were associated with the risk of prostate cancer in a U.S. population.

Methods: In a nested case-control design within the Health Professionals Follow-Up Study, we identified 700 participants with prostate cancer who were diagnosed after they had provided a blood specimen in 1993 and by January 31, 2000. Controls were 700 age-matched men without prostate cancer who had had a prostate-specific antigen test. We genotyped 19 common (>5%) haplotype-tagging SNPs chosen from the SNPs discovered in a resequencing study spanning TLR6, TLR1, and TLR10 to test for the association between sequence variants cluster and prostate cancer.

Results: Neither individual SNPs nor common haplotypes in the three gene regions were associated with altered risk of prostate cancer or subgroups of aggressive prostate cancer. No effect modification was observed for age, body mass index, or family history of prostate cancer, except that TLR6_3649 showed nominally significant interaction with family history at the P < 0.05 level.

Conclusion: Inherited sequence variants of the innate immune gene cluster TLR6-TLR1-TLR10 were not appreciably associated with the risk of prostate cancer in this cohort. (Cancer Epidemiol Biomarkers Prev 2007;16(10):1982–9)

Human Toll-like receptor (TLR) is a type I transmembrane protein with an extracellular domain consisting of a leucine-rich repeat region and an intracellular domain homologous to that of the Toll/interleukin-1 receptor (1, 2). TLRs play a role in the innate immune response and are involved in a pathway similar to that used by the Toll/interleukin-1 receptor. The human TLR family includes the TLR3, TLR4, TLR5, TLR2, and TLR9 subfamilies. The TLR2 subfamily is composed of TLR1, TLR2, TLR6, and TLR10 (3). Human TLR6, TLR1, and TLR10 are located on chromosome 4p spanning 54 kb and have one, four, and three exons, respectively. TLR1 and TLR6 are close to each other because they evolved from the same orthologous gene (3, 4). The amino acid sequence of human TLR6 shows 69% similarity with TLR1 (3). TLR6 is highly expressed in peripheral blood lymphocytes, the spleen, the thymus, the ovary, and the lung (4, 5). TLR1 is expressed in peripheral blood lymphocytes, epithelial cells, and endothelial cells (5, 6). TLR10 is highly expressed in the spleen (7) and B cells (8) and expressed at lower levels in the lungs, the small intestine, the stomach, the thymus, peripheral blood lymphocytes, lymph nodes, and the tonsils (5, 7, 9).

TLR1, TLR6, and TLR10 activate the nuclear factor-κB pathway (2). Additionally, TLR6 activates the c-Jun NH₂-terminal kinase pathway (4). TLR1 and TLR6 need to combine with TLR2 to form dimers to recognize pathogens (10, 11) and produce subsequent cytokine induction (12). The TLR1-TLR2 dimers recognize peptidoglycans from Gram-positive bacteria and zymosan from yeast cell walls; the TLR2-TLR6 dimers recognize the lipoproteins from Mycoplasma (11, 12). TLR10 can form either a homodimer or a heterodimer with TLR1 or TLR2; its cytoplasmic regions are also different from those of TLR1 and TLR6 (7). Genetic variations in TLR1 have been related to the risk of Lyme disease (10) and atherogenesis (13). TLR10 has been linked to asthma (14). Health effects related to sequence variants of TLR6 are unclear.

In TLR1-deficient mice, proinflammatory cytokine was not produced normally by macrophages when stimulated by lipoprotein and a synthetic triacylated lipopeptide (15). A murine model showed that a TLR2 and TLR6 heterodimer may be involved in immunomodulation of macrophages mediated by an antigenic protein, rLcrV (16). The Toll/interleukin-1 receptor domain of TLR10 is important in activation of promoters of certain inflammation cytokines (7).

Chronic inflammation has been associated with some cancers (e.g., cervix, stomach, primary liver, and bladder cancers; refs. 17, 18). Substantial evidence, including studies on sexually transmitted infections, clinical prostatitis, and genetic and circulating markers of inflammation and response to infection, supports a link between chronic intraprostatic inflammation and the risk of prostate cancer (19). Although TLR6, TLR1, and TLR10 are not expressed in prostate tissue, these genes are involved in the signaling pathway of pathogen-related innate immune responses, which may be involved in prostate carcinogenesis. A study of a Swedish population found that single nucleotide polymorphisms (SNP) and haplotypes in the TLR6-TLR1-TLR10 gene cluster were associated with the risk of prostate cancer (20). The same team later reported that the combination of IRAK4-7987 CG/CC and the risk genotype at the TLR6-TLR1-TLR10 gene cluster was associated with a 9.7-fold risk of prostate cancer compared with men with wild-type genotypes in the same Swedish population (21). Previous studies also showed significant associations between sequence variants of TLR4 and prostate cancer risk (22, 23). We hypothesized that genetic polymorphisms of the TLR6-TLR1-TLR10 gene cluster are associated with the risk of prostate cancer. Therefore, we explored the association between the TLR6-TLR1-TLR10 gene cluster and the risk of prostate cancer in the Health Professionals Follow-Up Study. SNP and haplotype analyses as well as case-only analyses were done to test this hypothesis.

This is a case-control study nested within the ongoing Health Professionals Follow-Up Study, in which incident prostate cancer cases were identified, with follow-up from 1986 to 2000. A total of 51,529 U.S. men aged 40 to 75 years were enrolled in 1986. Each participant completed a mailed questionnaire on demographics, lifestyle, and medical history and a semiquantitative food-frequency questionnaire at baseline. Information on exposures and diseases was updated every other year, and diet information was updated every 4 years. Deaths were identified through reports by family members, follow-up questionnaires, or a search of the National Death Index (24). This study was approved by the institutional review board at the Harvard School of Public Health. Completion of the self-administered questionnaire was considered to imply informed consent.

Between 1993 and 1995, blood samples from 18,018 of the participants were collected in tubes containing sodium EDTA. Samples were shipped by overnight courier and centrifuged; the aliquots, including plasma, erythrocytes, and buffy coat, were stored in liquid nitrogen freezers. We used a Qiagen QIAamp blood extraction kit for DNA extraction. All DNA samples were whole genome amplified, and the quality control samples had 100% genotype concordance rates. Among the men who gave a blood specimen, 95% responded to the year 2000 questionnaire; 18 died of prostate cancer before the end of follow-up and were included in the case series.

We identified 700 incident prostate cancer cases and 700 controls, which were composed of 94% self-reported Caucasians, 2.7% other races, and 3.5% without ethnicity data. To prevent the possible effect of population stratification, all analyses were restricted to Caucasians only (cases = 659, controls = 656). Each case was matched with one control who was alive, had not been diagnosed with cancer by the date of the case's diagnosis, and had a prostate-specific antigen (PSA) test after the date of blood draw. The latter criterion ensured that controls had the opportunity to have an occult prostate cancer diagnosed. All controls had a PSA test within 2.5 years of the date of diagnosis of their matched case. Cases and controls were matched on year of birth (±1 year), PSA test before blood draw (yes/no), and time (midnight to before 9 a.m., 9 a.m. to before noon, noon to before 4 p.m., and 4 p.m. to before midnight), season (winter, spring, summer, and fall), and exact year of blood draw because plasma analyses were being done on the same case-control set.

Haplotype-tagging SNPs were chosen using resequencing data from the Innate Immunity in Heart, Lung and Blood Disease-Programs for Genomic Applications. The Innate Immunity in Heart, Lung and Blood Disease-Programs for Genomic Applications resequenced the TLR6, TLR1, and TLR10 genes of 23 unrelated Europeans from Centre d'Etude du Polymorphisme Humain families, including 2.5 kb 5′ of the genes, exons, and 1.5 kb 3′ of the gene. Snp1 to Snp6 are common haplotype-tagging SNPs in TLR6 with frequencies >5%, chosen to tag haplotypes with frequencies >10% using the algorithm described in Sebastiani et al. (25). Snp7 was forced in because it is a nonsynonymous SNP. Snp9 and Snp10 are common haplotype-tagging SNPs in TLR1; Snp8, Snp11, and Snp12 were forced in for comparison with the previous study (20). Snp13 to Snp17 and Snp19 are common haplotype-tagging SNPs in TLR10, and Snp18 was forced in for comparison purposes as well. Laboratory personnel were blinded to case-control status. All case-control matched pairs were analyzed together using the Sequenom system. Multiplex PCRs were carried out to generate short PCR products (>100 bp) containing one SNP. The details of PCR and matrix-assisted laser desorption ionization time-of-flight mass spectrometry are available on request. Six control DNA samples were used for optimization. One SNP in TLR1 failed Sequenom assay design due to other SNPs that might interfere with primer annealing or extension. Another SNP in TLR1 was dropped after optimization because of high discordance rates among its duplicate samples. Finally, a total of 19 SNPs (Fig. 1; Table 1) were genotyped in seven plexes at the Harvard Partners Center for Genetics and Genomics (Boston, MA). For each SNP, genotyping data were missing in <5% of the study participants. Sixty-eight quality control samples were obtained from 18 external participants and each of them had two to six duplicates. These quality control samples were genotyped together with all other samples in this study. All quality control samples passed the quality control test (discordance rate = 0).

Investigators reviewed the medical and pathology records for men with prostate cancer reported from the follow-up questionnaire or death certificate to confirm adenocarcinoma of the prostate and to document clinical presentation, stage, and Gleason sum of the tumor. Because the concordance between self-reports and cases confirmed by medical record was high (>90%), self-reported cases for which we were unable to acquire medical records were included. The clinical presentation was categorized into elevated serum PSA concentration only, abnormal digital rectal examination with or without an elevated serum PSA concentration, or other/unknown. The cases were categorized into regionally invasive or metastatic (stage ≥T_3b, N₁, or M₁), organ-confined or minimal extraprostatic extension (T_1b to T_3a and N₀M₀), higher grade (Gleason sum ≥7), and lower grade (Gleason sum <7). Incidental microscopic focal tumors (stage T_1a) were excluded because they are generally indolent and susceptible to detection bias due to differential rates of surgery for benign prostatic hyperplasia. In addition, men with a previous cancer, except nonmelanoma skin cancer, before the date of blood draw were excluded. Confirmed non-T_1a tumors between blood draw and January 31, 2000 were included. In the blood subcohort, 92% of cases were confirmed by medical records and 5% by other corroborating information; only 3% were based on self-report (26).

The Hardy-Weinberg equilibrium (HWE) test was done for each SNP among controls. Haplotype block structure (Fig. 1) was determined by using Haploview⁵

and LocusView.⁶ The expectation-maximization algorithm was applied to estimate haplotype frequencies in each block using the tagSNP program (27). Conditional logistic regression models were used to estimate odds ratios (OR) for disease in participants carrying either one or two versus zero copies of the minor allele of each SNP and each multilocus haplotype; haplotype trend regression (28) was used to test global association between TLR6-TLR1-TLR10 haplotypes and prostate cancer. The type I error rate is controlled by the single multiple-degree-of-freedom test of association between TLR6-TLR1-TLR10 haplotypes and prostate cancer. Given a significant global test, haplotype- and SNP-specific tests can provide some guidance as to which variant(s) contributes to the significant global test, although the nominal P values we present do not control the family-wise error rate for these post hoc comparisons.

Age and family history are known risk factors for prostate cancer (29, 30); body mass index (BMI) is related to the risk of prostate cancer, although not consistently (31). We evaluated how these factors modified the association between TLR6-TLR1-TLR10 SNPs or haplotypes and the risk of prostate cancer by comparing a model with terms for main effects and interaction terms to a model with terms for main effects only using the likelihood ratio test. Because of the role of TLR6-TLR1-TLR10 in the innate immune response, the aggressiveness of prostate cancer may relate to genetic variations in the TLR6-TLR1-TLR10 gene cluster. We tested the association between TLR6-TLR1-TLR10 haplotypes and aggressiveness among patients with prostate cancer by using two definitions for tumor aggressiveness (aggressiveness 1: stage T_3b or T₄ or N₁ or M₁ or death due to prostate cancer; aggressiveness 2: stage T_3b or T₄ or N₁ or M₁ or death due to prostate cancer or Gleason sum ≥7). Aggressiveness 1 is useful in evaluating participants lacking information for Gleason sum and indicates how far a cancer has progressed independent of grade. Aggressiveness 2 indicates the potential of the tumor to progress by considering grade information. All analyses were conducted with Statistical Analysis System release 9.0 (SAS Institute), and all statistical tests were two sided.

Nineteen SNPs in TLR6-TLR1-TLR10 were genotyped. Among controls, Snp11 was out of HWE among controls (P = 0.002) and was therefore dropped from all the analyses (Table 1). Snp1, Snp4, Snp5, and Snp12 were out of HWE (P = 0.02-0.03), but we retained them in the analyses, as the less conservative Benjamini-Hochberg step-up procedure to control the false discovery rate (<5%) indicated that the null hypothesis of HWE was rejected only for Snp11. The internal blinded quality control specimens did not show evidence of genotyping error.

The study population included 659 incident prostate cancer cases and 656 matched controls. Age and BMI distributions were similar for cases and controls (Table 2). Family history of prostate cancer was significantly different between cases and controls (P = 0.009). The mean age at starting smoking, lifetime average number of cigarettes/day, and alcohol consumption were similar for cases and controls. Among cases, 78% were in tumor stage T_1b to T_3a, 73% had Gleason grade 5 to 7, 8% had aggressive prostate cancer defined by aggressiveness 1, and 36% had aggressive prostate cancer defined by aggressiveness 2. Eighteen cases died of prostate cancer before January 31, 2000.

All the individual SNP tests for associations were null (Table 3). Eighteen SNPs spanning TLR6-TLR1-TLR10 formed three linkage disequilibrium blocks using a modified version of the Gabriel et al. algorithm (32), where blocks identified with the default settings in Haploview are merged if they have multiallelic D′ >0.8, and the cumulative frequency of common (>5% frequency) haplotypes in the merged block is >80% (33). The first block included seven SNPs from TLR6, which constituted three common haplotypes (frequency >5%) with a cumulative frequency of 89.9% in controls (Table 4). The P value for the global test of the three common haplotypes was 0.62. The second block included three SNPs from TLR1, which constituted three common haplotypes (frequency >5%) with a cumulative frequency of 98.6% in controls (Table 4). The P value for the global test of the three common haplotypes was 0.50. Block 3 included eight SNPs, one SNP from TLR1 and seven SNPs from TLR10, which constitute three common haplotypes with a cumulative frequency of 91.7% in controls. The P value for the global test of the three common haplotypes was 0.59.

We also constructed the long-range 17-SNP haplotypes (without Snp4, Snp6, and Snp11) for comparison with the risk haplotype in Sun et al. (20), GATCTGCCGACTACCTG. Compared with the risk haplotype in Sun et al., ours lacked Snp11 because of its deviation from HWE. We found four common haplotypes constituted with these SNPs had a cumulative frequency of 77.1% (data not shown). No haplotypes were associated with risk of prostate cancer.

No SNPs showed nominally significant interaction with family history at the P < 0.01 level, and only one had a nominally significant interaction at the <0.05 level (Supplementary Table S1). BMI and age were not effect modifiers for the association between TLR6-TLR1-TLR10 SNPs and prostate cancer.

Case-only analysis (aggressive versus nonaggressive) was done among prostate cancer patients. No SNPs were associated with tumor aggressiveness (data not shown). Results were not significant for haplotype analyses as well (Table 5).

Chronic intraprostatic inflammation has been reported to increase the risk of prostate cancer (19). TLRs play an important role in pathogen-mediated innate immunity and chronic inflammation. TLR6, TLR1, and TLR10 belong to the TLR2 subfamily (3) and may influence carcinogenesis in the prostate by affecting the risk of chronic inflammation. In this study, we found no nominally significant (P < 0.05) associations with any haplotype-tagging SNPs or haplotypes in the TLR6-TLR1-TLR10 gene cluster and risk of prostate cancer. Prostate cancer family history modified the associations between Snp6 in TLR6 and the risk of prostate cancer (P = 0.03), but considering the large number of tests we conducted, the possibility that this result is a chance finding cannot be excluded.

Our results contrast with a Swedish study (20) that found that eight SNPs and six haplotypes in the same gene cluster were significantly associated with prostate cancer risk. None of the individual SNPs that were significantly associated with prostate cancer in the Swedish study were significantly associated in ours. Several factors could account for the disparity in results. Although both populations are Caucasian, the Swedish one is relatively homogeneous compared with ours. The case mix may have been different, as the cases in our population are heavily skewed toward early-stage PSA-detected cases. The Swedish study had about twice as many cases (including prevalent cases), and we may have been underpowered to confirm the associations the Swedish researchers observed. However, the direction of prostate cancer risk was opposite between the two studies and thus statistical power may not be an issue. In addition, there may have been some study design issues (e.g., SNP selection), although we believe that these were not major enough to account for the quite divergent results. Finally, if relevant, the TLR pathway would influence prostate cancer risk through some inflammatory pathway, and exposure to the presumed etiologic infectious agent(s) could differ in the two populations. Therefore, we believe that results from the Swedish study might not be generalizable to the U.S. Caucasian population.

Previous studies indicated that high BMI was related with low blood testosterone level and thus lower risk of early-onset prostate cancer (34). The other study, however, found that a low level of testosterone was related to high BMI (≥30), which reflects high insulin and insulin-like growth factor-I levels and, thus, high risk of high-grade prostate cancer (35). It is well known that the risk of prostate cancer significantly increases after age 50 (36, 37). However, the associations between SNPs in the TLR6-TLR1-TLR10 gene cluster and the risk of prostate cancer were not significantly modified by BMI and age in this study.

Our study had several limitations. Although population stratification cannot be excluded in a case-control study (38), our study population had a large sample size and was composed of 94% Caucasians. In addition, results from inclusion of all participants and all Caucasians were similar. This suggests that population stratification is not an issue in our population. For disease ascertainment, only 3% of cases were based on self-report. A high concordance rate (>90%) was found between self-report and medical record–confirmed cases. As a whole, these factors increase the validity of the results. An important limitation is that prostate cancer is a very heterogeneous cancer, and our case mix was largely composed of early-stage PSA-detected cancers. Although we had reasonable power to examine total prostate cancer as the end point, if an association exists for an important subgroup of cases (e.g., metastatic cancer), we were probably underpowered.

After recognition of microbial components, TLRs activate not only innate immunity but also adaptive immunity, largely by dendritic cells, which later express TLRs (e.g., TLR1, TLR2, TLR4, and TLR5; refs. 3, 39). TLR1 and TLR6 need to combine with TLR2 to form dimers to recognize pathogens (10, 11) and elicit subsequent cytokine induction (12). Studies on mice have shown that CD4TLR10 could interact with MyD88 and then activate nuclear factor-κB (2, 7). These mechanisms are very similar as TLR4 signaling pathway in response to pathogen recognition. However, data on human TLR10 are limited. A previous study (22) shows some evidence of association, supporting the contention that variation in TLRs may reduce or even block the signaling of the immune response (e.g., chronic inflammation) and thereby lower the risk of prostate cancer. But this study did not find any association with TLR6-TLR1-TLR10. Combined with the previous study, the results are mixed. The inconsistent results with the TLR6-TLR1-TLR10 gene cluster as well as TLR4 between the Swedish study and ours suggest the problem of low reproducibility across genetic studies. Because of the limited amount of research on inflammation and prostate cancer, more studies on these genes will be needed to confirm these findings. Future studies related to biological pathways may help us understand the mechanisms of prostate cancer risk.

We thank Monica Coleman for her assistance; Pati Soule and Ana-Tereza Andrade for DNA sample extraction; and Dr. David Kwiatkowski, Alison Brown, and Maura Regan (Partners High-Throughput Genotyping Center) for genotyping.