A Large Prospective Study of SEP15 Genetic Variation, Interaction with Plasma Selenium Levels, and Prostate Cancer Risk and Survival

Read the Perspective on this article by Platz, p. 576

Several prospective studies, including our previous reports (1, 2), have suggested that selenium may act as a chemopreventive agent in prostate cancer (PCa). The Nutritional Prevention of Cancer Trial found a 52% decrease in PCa incidence after 13 years of follow-up among those randomized to 200 mg of selenium per day (3, 4). This trial prompted the initiation of the Selenium and Vitamin E Cancer Prevention Trial (SELECT; ref. 5), which was halted early due to nonsignificant increases of PCa risk in the vitamin E arm and diabetes in the selenium arm, limiting conclusions that can be made on the long-term effect of selenium on risk of advanced PCa and PCa-specific mortality. Prospective studies of blood and toenail levels of selenium have suggested inverse associations (6), particularly for “aggressive” cancer, defined as Gleason score ≥7 or extraprostatic tumors (1, 2). A recent meta-analysis for selenium and PCa pooled risk estimates from 11 cohort studies to obtain a relative risk of 0.72 (0.61-0.84) comparing higher to lower levels of intake (7). However, the role of selenium in PCa progression or PCa-specific mortality has not been fully elucidated.

One potential biological mechanism might be through the antioxidant function of selenium. A variant in the antioxidant enzyme manganese superoxide dismutase (MnSOD) gene may modify the effect of selenium on the risk of developing PCa (8) or the aggressiveness of disease (9). In addition, selenium is incorporated into several antioxidant proteins, including the glutathione peroxidases and the thioredoxin reductases, and is necessary for their redox function. Genes that encode selenoproteins contain a selenocysteine insertion sequence structural element in the 3′ untranslated region of the mRNA. The selenocysteine insertion sequence element causes the UGA codon, normally the stop codon, to encode a selenocysteine, which is necessary for the structure and function of the resulting protein and often has redox properties.

A recently discovered selenoprotein, 15-kDa selenoprotein (SEP15), is highly expressed in the prostate (10). Although its function is unknown, SEP15 binds to UDP-glucose:glycoprotein glucosyltransferase (11–14), a protein that regulates the quality control pathway that assists in the folding of N-linked glycoproteins in the endoplasmic reticulum, which suggests that SEP15 may function in this pathway. SEP15 may have antioxidant properties as well (15). A single-nucleotide polymorphism (SNP) in the 3′ untranslated region of SEP15 (rs5859, 1125 G/A) was found to decrease the efficiency of the selenocysteine insertion sequence element at higher concentrations of selenium (16). Additionally, malignant mesothelioma cell lines with the A allele were less responsive to the growth-inhibiting and apoptotic effects of added selenium than those with the homozygous G genotype (17); high selenium levels may therefore be necessary for those with the A allele. In a case-control study, this same polymorphism modified the association between selenium status and lung cancer risk in smokers; for G allele carriers, higher selenium levels increased the risk of lung cancer, whereas in those with the AA genotype, high selenium levels seemed protective (18).

The role of SEP15 genetic variants in the development or progression of PCa, either independently or as modifiers for the effects of selenium, is unknown. Therefore, we examined the association of polymorphisms in SEP15 with PCa risk and mortality in the Physicians' Health Study (PHS).

The PHS began as a randomized, double-blind, placebo-controlled trial investigating the role of aspirin and β-carotene in the prevention of cardiovascular disease and cancer among 22,071 healthy U.S. physicians. Men were excluded if they had any serious medical conditions including all cancers (except nonmelanoma skin cancer). Blood samples were collected from 68% of the physicians at baseline in 1982-1984. A detailed description of the PHS has been published previously (19). Participants are followed through annual questionnaires to collect data on diet, health and lifestyle behaviors, and medical history, and biannually through postcards to ascertain health end points, including PCa. All self-reported PCa cases are verified through medical record and pathology review. Through this systematic medical record review, we also abstract data on PSA at diagnosis, tumor stage, and Gleason score. Death certificates and records are obtained and reviewed to determine cause of death. The follow-up rate for cancer incidence is 96% and that for mortality is 98%. Metastases are reported on follow-up questionnaires sent to all men living with PCa and confirmed by medical record review.

To assess associations with PCa incidence, we used a nested case-control study, with controls selected through risk-set sampling and matched to cases on age at baseline (±1 year for cases ≤55 and, if necessary, within 5 years for older cases), smoking status (never, former, current), and follow-up time. For the current study, we restricted participants to self-reported Caucasians to reduce potential population stratification. For the case-control analysis, 1,286 cases (diagnosed from 1982 to 2005) and 1,267 controls were included; an additional 45 PCa cases that were diagnosed after they were matched as controls in the case-control analysis were included for the outcomes analyses (n = 1,331). During follow-up through March 1, 2008, 178 men died of PCa or developed distant metastases.

Using the HapMap database, we selected four SNPs to capture variation (with R² > 0.80) within SEP15 and 5 kb upstream and downstream (total 62 kb). Selection was restricted to SNPs with a minor allele frequency >5% in the International HapMap CEPH samples (n = 37) and performed using the Tagger Pairwise program (http://www.hapmap.org). We also included rs5859 (G1125A) because it had previously been reported as functional. DNA was extracted from whole blood. Genotyping was done with Sequenom iPLEX matrix-assisted laser desorption/ionization-time of flight mass spectrometry technology. All SNPs had >91% genotype completion rates.

In a subgroup of participants, plasma levels of selenium (799 cases in the survival analysis; 782 cases and 540 controls in the risk analysis), α-tocopherol, and lycopene (971 cases in the survival analysis; 938 cases and 830 controls in the risk analysis) were measured as previously described (2, 8, 20). Each sample was tested in duplicate; the mean coefficient of variation for all analytes ranged from a low of 6.4% for selenium to a high of 11.9% for α-tocopherol.

Analyses were done with SAS version 9.1 statistical software; all P values were two-sided. Using Pearson's goodness-of-fit test, none of the SNPs violated Hardy-Weinberg equilibrium (P > 0.05). SNPs were analyzed under a codominant genetic model, as well as a recessive model for the survival analysis and interactions.

In a nested case-control study, we assessed the risk of incident PCa using unconditional logistic regression models, adjusting for the matching factors (age at randomization, smoking status, and duration of follow-up). We also conducted a subgroup analysis, comparing aggressive cases (defined as Gleason score ≥8, clinical stage T₃, T₄, N₁, or M₁, or fatal/metastatic disease) to controls. Among PCa cases, we performed an analysis of time to lethal PCa outcome using a Cox regression model adjusting for age at diagnosis. A lethal PCa outcome was defined as death due to PCa (n = 161) or the development of bony metastases (n = 17); follow-up began at the time of PCa diagnosis and individuals were censored at the time of death from another cause or the end of follow-up (March 1, 2008).

Haplotypes and frequencies for cases and controls were created using the Haploview program (21). Only one haplotype carried the risk allele for rs1407131, and it also carried the risk alleles for the other significant SNPs (ATACG); therefore, individuals who are homozygote for all three risk alleles carried two copies of this haplotype. These individuals were compared with all others for PCa incidence (with unconditional logistic regression) and mortality (with Cox regression model). To assess the independent association of the SEP15 haplotype with cancer mortality, we further adjusted for Gleason score and clinical stage.

Prediagnostic plasma selenium was normally distributed. We used linear regression to determine if the SEP15 SNPs were associated with selenium levels among cases and controls combined, adjusting for analytic batch, the matching factors, and case-control status. Quartiles of selenium were created using batch-specific cut points from the controls in the case-control analysis and batch-specific cut points from all cases for the mortality analysis. Interaction terms of ordinal quartiles of selenium and genotype were used in an unconditional logistic regression model for risk and Cox regression model for mortality. We created an antioxidant score for plasma selenium, lycopene, and α-tocopherol by summing the quartiles of these three biomarkers, as described previously (2). Four score categories were created (3–12); the interaction of this variable and genotype with risk and mortality was tested as described above for selenium. A score variable that included only the quartiles of lycopene and α-tocopherol was created to determine if any interactions were specific for selenium (2–8).

Characteristics of the PHS PCa cases and controls are presented in Table 1. Figure 1 shows the relative positions and linkage disequilibrium (LD) across the five SEP15 SNPs. Genotype frequencies were similar across cases and controls (Table 2).

We observed no statistically significant associations under an additive model for the SEP15 SNPs with risk of PCa or aggressive PCa in the nested case-control analysis (Table 2). Under a recessive model, rs527281 was significantly associated with risk, although the very small number of individuals with the minor allele homozygote genotype led to wide confidence intervals [odds ratio, 6.34; 95% confidence interval (95% CI), 1.43-28.13]. However, survival analysis among PCa cases revealed that three of the SNPs were significantly associated with PCa mortality (Table 3). After testing the codominant model, for rs479341 and rs1407131 we observed what seemed to be a recessive association and therefore tested the recessive model; the less frequent homozygote genotype was associated with increased risk of PCa mortality [hazard ratios (HR) of 1.94 (95% CI, 1.15-3.25) and 2.85 (95% CI, 1.45-5.59), respectively]. SNP rs561104 also followed a recessive pattern; the AA homozygote of rs561104 showed an increase in the risk of mortality when compared with the reference of AG/GG (HR, 1.54; 95% CI, 1.12-2.11); additionally, there was a statistically significant trend for rs561104 with the codominant model (P_trend = 0.02). SNP rs5859, previously identified as functional and in strong LD with rs479341, was associated with a nonsignificant increased risk of PCa-specific mortality (HR, 1.68; 95% CI, 0.91-3.11; P = 0.10).

Because the three significant markers were somewhat correlated (R² 0.10-0.59; Fig. 1), the significant associations of these three SNPs with PCa mortality may not be completely independent. When all three SNPs, modeled as recessive, are included in a model together, only rs561104 remained significantly associated with survival (P = 0.03); however, this SNP is the most common and has the smallest correlation with the other two SNPs. We considered that perhaps there were multiplicative effects, where having more than one risk allele (carried on a particular haplotype) may have been worse than only carrying one risk allele. Using the Haploview software (21), we found that only one haplotype carried the risk allele for rs1407131; this haplotype additionally carries the risk allele for rs479341 and rs561104 (ATACG). Therefore, individuals with two copies of each risk allele carry two copies of this haplotype. The frequency of this haplotype was similar in cases (13.9%) and controls (13.7%) and was unrelated to risk of incident PCa (data not shown). However, as was seen in the analysis for rs1407131, cases carrying two copies of the risk haplotype had a 3-fold increased risk of PCa-specific mortality compared with all other cases (HR, 3.07; 95% CI, 1.56-6.02). This result remained significant even after adjusting for Gleason score (HR, 2.60; 95% CI, 1.06-6.38) or clinical stage (HR, 2.09; 95% CI, 1.01-4.31). The three individual SNPs, rs561104, rs1407131, and rs479341, remained significant on their own after adjusting for clinical stage [HR, 1.44 (P = 0.03); HR, 1.80 (P = 0.04); HR, 2.08 (P = 0.05), respectively], but not for Gleason score [HR, 1.31 (P = 0.15); HR, 1.63 (P = 0.14); HR, 2.08 (P = 0.11), respectively].

SNPs in SEP15 could alter its function or expression, modifying the effect of selenium on PCa risk or progression. None of the five SNPs was significantly related to plasma selenium levels and there were no significant interactions between genotypes and selenium for risk (results not shown). Prediagnostic selenium levels were not independently associated with overall PCa survival (HR for extreme quartiles, 0.88; 95% CI, 0.54-1.41; P = 0.58). However, we found a significant interaction between rs561104 genotypes and quartiles of selenium levels for PCa-specific mortality (P_interaction = 0.02). An inverse association with quartiles of selenium was apparent only among individuals carrying the minor G allele (HR, 0.82; 95% CI, 0.67-1.00; P_trend = 0.05), whereas the beneficial association was completely absent in those with the homozygous variant AA genotype (HR, 1.08; 95% CI, 0.85-1.38; P_trend = 0.53; Table 4).

We also created a variable for antioxidant score based on quartiles of selenium, lycopene, and α-tocopherol, as shown in our previous publication (8); the significance of the interaction of score with rs561104 was attenuated (P_interaction = 0.14) compared with the interaction with selenium alone. When we excluded selenium from the score, no significant interaction between the antioxidant score categories and SNPs was observed for mortality (P_interaction = 0.39), suggesting a selenium-specific effect of SEP15 on PCa mortality, rather than a general antioxidant effect.

In this long-term prospective study, we observed no significant associations of five SNPs in the selenoprotein SEP15 with the incidence of PCa. However, genetic variation in SEP15 was significantly associated with PCa survival. The SEP15 variants, and therefore selenium, may be more important later in the progression of PCa than in the initiation of disease. The “risk” allele for rs1407131 fell on only one haplotype, which also carried the risk alleles for the other two SNPs significantly associated with PCa-specific mortality; cases carrying two copies of this haplotype had a 3-fold increased risk of dying from PCa compared with all other PCa cases (P = 0.001). This result remained statistically significant even when adjusting for Gleason score and clinical stage, the strongest predictors of PCa mortality, suggesting potential independent predictive value of the SEP15 gene in PCa progression. However, we recognize that such adjustment likely represents over-control for the process of cancer progression, as the progression to lethal disease acts through more advanced grade and stage. We therefore present the age-adjusted result as the main estimate of the magnitude of the effect.

These results suggest that variation in SEP15 is involved with the progression of PCa, but whether this is through a protein folding function, antioxidant properties, or some other pathway remains unknown. In this study, we found that, among the cases with two copies of the rs561104 A risk allele, plasma selenium levels were unrelated to PCa-specific mortality. However, among carriers of the common G allele, higher selenium levels were associated with a significant lower risk of PCa mortality. Because G allele carriers are common (64% in controls and 67.5% in cases), we expect that perhaps two thirds of the men with PCa could potentially benefit from selenium in delaying cancer progression. It is possible that the variant allele, or a causal marker in LD with this SNP, prevents the incorporation of selenium into SEP15. If SEP15 is not functioning properly, given its role in protein-folding quality control, other proteins may not function properly, leading to additional damage within the cancer cells and disease progression.

Our previous analysis for risk of aggressive PCa showed a stronger interaction of a variant in the MnSOD gene with plasma antioxidant score (combining selenium, lycopene, and vitamin E) than with plasma selenium alone (8). However, we found that the significant interaction of this SEP15 variant was exclusively with selenium and not with other antioxidants. This may indicate that the effect of the SNP or the haplotype on mortality is due to relative insensitivity to selenium.

Although we cannot exclude the possibility that the observed associations are attributable to other variants in LD with these markers, the specific interaction of the SEP15 variant with selenium increases the plausibility of the association. It is also possible that the associations we observed are false positives. However, in our primary survival analysis, after performing a conservative Bonferroni adjustment for 10 codominant and recessive tests, the recessive model results are still marginally significant (rs479341 P = 0.10; rs561104 P = 0.08; rs1407131 P = 0.02).

Strengths of this study include its large sample size, prospectively collected plasma samples, and long duration of follow-up with cancer death and metastases as the outcome. Using a tag SNP approach, we were able to comprehensively examine common variation in the SEP15 gene. Having antioxidant biomarker data and genetic information on the same individuals allowed us to attempt to elucidate the role of these SNPs in PCa progression.

This study provides novel evidence that SEP15 may play an important role in PCa survival. The association of prediagnostic selenium levels with PCa survival was significantly modified by a SEP15 SNP, implying that some PCa cases may benefit from selenium more than others, depending on their genetic variants. Further work will be done to replicate these findings and explore functionality; it will be of interest in the SELECT trial to determine if selenium had a protective effect on disease progression in any genetic subset of participants (22).

T. Kurth has received within the last 2 years investigator-initiated research funding as principal or co-investigator from the NIH and Merck. Further, he is a consultant to i3 Drug Safety and World Health Information Science Consultants, LLC; he has received honoraria from Genzyme, Merck, and Pfizer for educational lectures. The other authors disclosed no potential conflicts of interest.

Grant Support: Department of Defense grants PC050569 and PC073618. The PHS was supported by grants CA-42182, CA-34944, CA-40360, and CA-097193 from the National Cancer Institute and grants HL-26490 and HL-34595 from the National Heart, Lung, and Blood Institute, Bethesda, Maryland. K.L. Penney was supported by National Research Service Awards T32 CA009001-32 and R25 CA098566.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.