|Home | About | Journals | Submit | Contact Us | Français|
Lipid peroxidation is considered a unifying mechanistic pathway through which known risk factors induce renal cell carcinoma (RCC). We hypothesized that genes selected apriori for their role in lipid peroxidation would modify cancer risk. We genotyped 635 single nucleotide polymorphisms (SNPs) in thirty-eight candidate genes in 777 Caucasian RCC cases and 1035 controls enrolled in a large European case-control study. Top candidate SNPs were confirmed among 718 Caucasian cases and 615 controls in a second study in the United States. Two of the three SNPs (rs8106822 and rs405509) that replicated in the US study were within a regulatory region of the APOE promoter. The odds ratio (OR) for rs8106822 A>G variant was 1.22AG and 1.41GG (p-trend=0.01) in the European study, 1.05AG and 1.51GG (p-trend=0.03) in the US study, and 1.15AG and 1.44GG (p-trend=0.001) among 1485 cases and 1639 controls combined. The rs405509 G>T variant was associated with risk in the European (OR=0.87TG; OR=0.71TT; p-trend=0.02), the US (OR=0.68TG; OR=0.71TT; p-trend=0.02), and both studies combined (ORTG=0.79; ORTT= 0.71; p-trend=0.001), as was the G-G haplotype (r2=0.64; p=4.7 × 10-4). This association is biologically plausible as SNP rs405509 was shown to modify protein binding and transcriptional activity of the APOE gene in vitro and is in LD with key known variants defining the e2, e3, e4 alleles that modify risk of atherosclerosis, Alzheimer's disease risk, and progression to AIDS. In two large case-control studies, our findings further define a functional region of interest at the APOE locus that increases RCC susceptibility.
The association between kidney cancer and known risk factors such as obesity and hypertension, tobacco use, and suspected occupational risk factors such as chlorinated solvents, gasoline, and lead has been supported through several epidemiologic studies (1-4). In humans, evidence that oxidative damage and lipid peroxidation may be important intermediate risk factors underlying kidney carcinogenesis has come from studies reporting higher rates of lipid degradation by-products among cancer cases with the above known risk factors, and related risk factors such as diabetes (5-7). In animal models, by-products that result from lipid peroxidation of the renal tubules can form DNA adducts, leading to alterations relevant to carcinogenesis (8,9). Lipid peroxidation by-products can also cause direct oxidative damage at the DNA, protein, and cellular levels.
To further clarify the role of lipid peroxidation in kidney cancer and specifically renal cell carcinoma (RCC) the most common form of kidney cancer, we selected a priori 38 candidate genes for analysis with 635 tagging single-nucleotide polymorphisms (SNPs) that provided comprehensive genomic coverage of each candidate gene and regulatory regions upstream and downstream from coding regions among subjects enrolled in a large hospital-based case-control study of kidney cancer. This study was sufficiently powered to investigate modification of cancer risk associated with common genetic variation. To confirm these findings, we had the opportunity to select three promising markers for rapid replication among cases and frequency-matched population controls from a kidney cancer case-control study conducted in the United States. In total, this study included 1485 genotyped cases and 1639 controls.
This study is a hospital-based case-control study that was conducted in seven centers in Central and Eastern Europe (Moscow, Russia; Bucharest, Romania; Lodz, Poland; and Prague, Olomouc, Ceske-Budejovice and Brno, Czech Republic). Details of the study have been described previously (10). Patients with newly diagnosed and histologically confirmed kidney cancer (ICD-0-2 code C64) between the ages of 20 and 79 years were recruited from August 1999 through January 2003. Trained medical staff reviewed medical records and extracted information on date and method of cancer diagnosis, histological classification and confirmation of the RCC subtype, tumor location, stage and grade. Eligible controls were chosen from among patients admitted to the same hospital as cases for conditions unrelated to smoking or genitourinary disorders (except for benign prostatic hyperplasia) and were frequency-matched to cases on age (within 3 years), sex, and study center. Some controls had been previously recruited from an earlier case-control study of lung and head and neck cancer (11). No single disease made up more than 20% of the control group. Response rates at each center ranged from 90.0 to 98.6% for cases and from 90.3 to 96.1% for controls. Interviews were conducted by trained personnel to collect data on demographic characteristics, education, tobacco smoke exposures, alcohol consumption, dietary practices, medical history, family history, and occupational history. In total, 1097 cases and 1476 controls were interviewed. Blood samples were collected and stored at -80°C. Genomic DNA was extracted from whole blood buffy coat by the standard phenol chloroform method at an NCI designated laboratory from 987 of 1097 (90%) of cases and 1298 of 1476 (88%) of controls. All subjects in this study provided written informed consent. This study was approved by the institutional review boards of all participating centers.
This study is a population based case-control study conducted in Detroit, Michigan and Chicago, Illinois, in the United States. Cases included residents of each study area aged 20-79 years who were newly diagnosed with histologically confirmed renal cell carcinoma (ICD-02 C64.9) from February 2002 through January 2007. Controls were frequency-matched to cases by study center, race, age, and sex. Controls aged 65 years and older were identified from Medicare files, and those under age 65 years were identified from Division of Motor Vehicle records. Written informed consent was obtained from all participants, and institutional review board approvals were obtained from all participating study centers. Participants were interviewed by trained interviewers to collect information on demographic background, consumption of tobacco, alcohol and other diet, occupational history, as well as known or suspected risk factors for kidney cancer such as hypertension and antihypertensive medicines. Buccal and blood samples were collected as a source of genomic DNA. A total of 1568 Caucasians (856 cases and 712 controls) were interviewed. Of these subjects, blood samples were collected from 718 of 856 (83.9%) of cases and 615 of 712 (86.4%) controls. Genomic DNA was extracted using standard phenol chloroform methods at an NCI designated laboratory, and 708 Caucasian cases and 604 controls were successfully genotyped.
We analyzed 635 tagging SNPs spanning regions both upstream and downstream of 38 candidate genes involved in mediation and response to lipid regulation, peroxidation, and metabolism: [gene (number of tagging SNPs)]: AKRFC3 (13), ALOX5 (8), ALOX12 (8), APOB (14), APOE/C1 (5), ATP1A2 (17), CAT(17), COMT(34), CYP17(9), CYP19(35), FOXO1(17), FOXO3 (9), GGH (7), GPX1(3), GPX2 (10), GPX3(15), GPX4 (5), GSR(9), GSTO1 (7), HOA2 (10), IL12A (16), IL12B (16), INS(12), INSR(49), LEPR(27), MGMT(57,) MUTYH(8) NOS2A(19), NOS3(11), NOX1(8), PON 1/2/3(48), PPARA(22), PPARG(9), PTGS(7), SOD2(8), SOD3(8), TXN(22)) (Suppl T1). SNPs were selected to provide high genomic coverage. Tag SNPs were selected 20 kb 5′ of the start of transcription and 10 kb 3′ of the last exon using HapMap CEU data (http://www.hapmap.org) among SNPs with minor allele frequencies of at least 5% and an r2 ≥0.80 (12). In addition, nonsynonymous SNPs or those correlated with polymorphisms with potential functional significance were included in the analysis. All SNPs and assay information are reported in the NCI SNP500 Cancer database (http://snp500cancer.nci.nih.gov) (13). Genotyping was conducted at NCI's Core Genotyping Facility where staff was blinded to case/control status. Duplicate quality control samples (5% samples) were interspersed among plates. All genotyping was performed using an Illumina GoldenGate ® Oligo Pool All (OPA) assay, which was designed using publicly available sequencing information. The genotype frequencies among controls showed no deviation from the expected Hardy-Weinberg equilibrium proportions (p > 0.05). The genotyping completion rate ranged from 98-100% for all SNPs, except was 97% for rs2240714 tagging TXNRD2/COMT, rs4135182 tagging TXN gene region, and rs732498 tagging the 3′ region of the SOD2 gene.
For the replication study, TaqMan based assays were used. Methods for TaqMan based assays for rs8106922, rs405509, and rs4795067 can be found at http://snp500cancer.nci.nih.gov/home.cfm. Completion rates for each of the SNPs was >99%. Concordance rates were >98%.
Initial exploratory analyses were based on the 777 cases and 1035 controls from the Central and Eastern European study that provided a sufficient quantity and suitable quality of genomic DNA for genotyping on the Illumina GoldenGate ® platform. Associations were evaluated through a comprehensive sequence of methods that were sensitive to detect significant associations by gene and by region, while controlling for type I error that is inherent in a large study of 38 genes and 635 SNPs. First, subject characteristics having statistically significant (p<0.05) differences in distributions between cases and controls were examined for their associations with SNPs and RCC risk using a chi-square test, and were evaluated as potential confounders. Genotypes were evaluated by coding the homozygous common allele as the referent group and separately comparing the heterozygous and homozygous rare allele genotypes to the most common homozygous referent group. Linear tests for trends were conducted by including a variable coded 0, 1, and 2 corresponding to the number of rare alleles. Unadjusted and adjusted (age, sex, and study center, smoking, BMI, hypertension) single marker odds ratios (OR) and 95% confidence intervals (95% CI) were generated both for single SNPs and SNPs in regions with a high level of signal. Linkage disequilibrium (LD) between markers was assessed by calculating pairwise Lewontin's D′ and r2 values using the Haploview program among controls applied to genotype data.
For gene based analyses, global p-values were evaluated using the minimum-p value permutation test (MIN-P). This method corrects for multiple testing while also accounting for correlations between SNPs within a gene (14). In addition, a Haplowalk sliding window analysis of three consecutive SNPs was used to identify chromosome regions of interest that remained significant at a false discovery rate (FDR) level of 1% or less; FDR is defined as the expected proportion of falsely rejected hypotheses among the rejected hypotheses (15,16). Given k genotyped SNPs in a gene, the Haplowalk procedure considered a 3-SNP window across each gene from SNP 2 through SNP K-1, resulting in K-2 analyzed overlapping windows. For each window, haplotype frequencies in cases and controls were reconstructed using the EM algorithm, and a Wald test was used to screen for association with case-control status (15,16). The Benjamini-Hochberg FDR procedure was applied to the set of k-2 Wald-test p-values yielding FDR adjusted p-values that control for multiple testing at the gene level (16). In the initial screening phase, no adjustment was made for potential confounders, and the Wald tests used a threshold value of 5%, such that haplotypes in cases and controls with an estimated frequency below the threshold value in controls were combined into a separate “rare haplotype” category for purposes of statistical testing. Genes with a MIN-P value less than or equal to 0.05, and genes in which we identified one or more region(s) using the haplowalk analysis, and with an FDR adjusted p-value less than or equal to 0.05, were reconstructed and associations evaluated using Haplostats (Version 1.3.1) in R (version 2.4.1) adjusting for sex, age, center, smoking status, BMI, and hypertension, using the most common haplotype as the referent.
The replication analysis was conducted among 708 Caucasian cases and 604 controls from the US study. Three SNPs; rs8106922, rs405509, and rs4795067 were selected for replication based on having high minor allele frequencies (between 30-50%) to increase study power to detect the odds ratios observed in the European study. The same statistical methods were used where applicable, without adjustment for multiple comparisons. Therefore, estimates were calculated for both studies independently and combined and adjusted for center, sex, age in quintiles based on combined controls. Additional adjustment for potential confounders (body mass index (BMI), self-reported hypertension, and smoking) did not result in meaningful changes in risk estimates and therefore were not included in models. In addition, we investigated multiplicative interactions between individual SNPs and age, sex, BMI, self-reported hypertension, and smoking using the likelihood ratio test to compare the fit of models with and without interaction terms but we did not find evidence of heterogeneity.
Heterogeneity of genotype frequencies between countries was evaluated by using the likelihood ratio test to compare the fit of models with and without interaction terms. We found no evidence of heterogeneity across study centers. Moreover, no evidence of population stratification was apparent from a principal components analysis of a genome wide association study conducted in this population (17), and the likelihood of this is small among European populations (18).
Among cases and controls in the European study, most of the study population was from the Czech Republic, and a slightly higher proportion of cases were from that country (Table 1). Controls were more likely to be male, but were similar in age distribution. The prevalence of subjects in the lowest (<51 years) and highest age quintiles (>=69 years) were about twice as high in the European study than in the US study. Cases were more likely than controls to have higher BMI, and self-reported hypertension (4). The prevalence of clear cell renal cancer was 83% in both studies.
Ranked results from global gene-based tests of association (MIN-P test), the false discovery rate (FDR)-adjusted minimum p-trend for the additive models, and the smallest FDR-adjusted p-value from the 3-SNP sliding window analysis are presented in Table 2. Based on the MIN-P test, four genes were selected for in-depth analysis after multiple testing correction of single marker associations: APOE, GPX4, NOS2A, PTGS2, The overall MIN-P adjusted p-values for these were 0.017, 0.020, 0.055, 0.069 respectively. These genes also had the lowest minimum p-values from a 3-SNP sliding window analysis with a false discover rate (FDR) level<1% (4.8×10-4, 7.4×10-4, and 1.7×10-4, 1.09×10-5, respectively). Individual associations between tagging SNPs and risk associated with these four genes are presented in Suppl Table 3. For the APOE/C1 gene region[see supplemental Figure 1 to see correlations between APOE gene tagging SNPs in haploview], significant inverse associations were observed with two promoter region SNPs, one rare variant located in the APOC1 gene promoter (rs283813) and the other in the APOE gene promoter region (rs405509). In addition, increased risk was observed with the G allele of SNP rs8106922. For the GPX4/POLR2E region, increased risk was observed for one of the five tagging SNPs (rs11668388). For the NOS2A gene, five of 19 SNPs significantly modified RCC risk. Two SNPs were located in the promoter region (rs2531860, rs2779248), and three were intronic (rs3729508; rs4795067; rs2248814). Three of five PTGS2 tag SNPs were also significantly associated with risk, more than would be expected by chance.
Subsequently, genomic regions identified from the sliding window analysis were reevaluated after adjustment for potential confounders in R using Haplostats (Suppl Table 4). Two variant haplotypes in the APOC1 regulatory region were inversely associated with risk (T-T-A, OR=0.76; 95% CI:0.61-0.94; p=0.01) and A-C-A (OR=0.56; 95% CI:0.40-0.79; p=9×10-4) when compared to the common referent haplotype T-C-A (P-global=0.002). This association was similar in magnitude to that observed for the rare APOC1 (rs283813) AA variant that was found in only 3% of cases and 8% of controls (OR=0.69TA, OR=0.51AA; p-trend =0.004). In contrast, when SNPs (rs8106822, rs405509) tagging the APOE promoter region were considered as a haplotype, increased risk was observed for the G-G haplotype compared to the common haplotype A-T (OR=1.21; 95% CI:1.05-1.40, p=0.01). Analysis of the GPX4 gene region revealed increased risk associated with the haplotype A-G-A compared to the common referent T-A-A haplotype (OR=1.22; 95% CI:1.05-1.42, p=0.01), although the rs11668388 variant appeared to be driving the association (data not shown). Additional adjusted haplotype analysis of the NOS2A and PTGS2 gene regions that were significantly associated with risk are presented in Suppl Table 4.
Subsequently, three loci (rs8106922, rs405509, and rs4795067) were selected for rapid replication within regions of interest having high minor allele frequencies (between 30-50% to increase study power). SNPs were analyzed among 708 RCC cases and 604 frequency-matched population Caucasian controls from the US Renal Cancer Study (Table 1). SNPs rs8106922 and rs405509 in the APOE promoter region were significantly associated with risk in the US study (Table 3). As observed in the European Study (OR=1.22AG; OR=1.41GG; p-trend=0.01), increased risk was associated with the G allele of rs8106922 in the US study (OR=1.05AG; OR=1.51GG; p-trend=0.03), and when the two studies were combined among 1485 cases and 1639 controls (OR=1.15AG; 1.44GG; p-trend=0.001). Similarly, as observed in the European study (OR=0.87TG; OR=0.71TT; p-trend=0.02), decreased risk was associated with the T allele of rs405509 in the US study (OR=0.68TG; OR=0.71TT; p-trend=0.02). When data from the two studies were combined, the results for rs405509 were strengthened (ORTG=0.79; ORTT=0.71; p-trend=0.001), as were the results from the combined G-G haplotype (r2=0.64) when compared to the A-T referent (OR=1.22; 95% CI: 1.09-1.36, p=4.7×10-4; p-global=0.003). The NOS2A SNP rs4795067 that was selected for replication based on the results from the European study (ORTC=1.33; 95% CI:1.08-1.63, ORTT=1.50; 95% CI:1.11-2.04, p-trend=0.002) was not significantly associated with risk in the US study (ORTC=1.05; 95% CI:0.83-1.32, ORTT=1.15; 95% CI:0.79-1.67, p-trend=0.48). When genotyping data from the two studies were combined, risk was significantly elevated among subjects with at least one variant allele (ORTC=1.17; 95% CI:1.01-1.37 and ORCC=1.34; 95% CI:1.06-1.37; p-trend=0.006). Results were similar when analyses were restricted to include clear cell renal cancer cases only (Table 3). No interactions between our significant SNPs and potential and risk factors/environmental exposures were detected (data not shown).
Although still relatively rare, RCC incidence has increased rapidly in the U.S. over the past few decades (19-21). Several well-established lifestyle risk factors such as BMI, hypertension, and smoking have been identified and are thought to explain approximately 50% of cases (3). Causes for the remaining half of cases remain unknown. Genetic susceptibility to sporadic kidney cancer is a promising area of research, and has not been fully investigated. To our knowledge, this is the first evaluation of variation in lipid metabolism/peroxidation genes in the two largest kidney cancer case-control studies with genomic DNA conducted to date. These findings are important as lipid peroxidation has emerged as a unifying mechanism through which several known and suspected risk factors are thought to modify kidney cancer risk. Many kidney cancer genetic susceptibility studies conducted to date have been under-powered, and have focused primarily upon common variation in xenobiotic metabolism (reviewed in 20,21), micronutrient metabolism (20,21,22), in addition to genes associated with familial cancers such as von Hippel-Lindau (VHL) (20,21) and did not employ a tagging approach across genes selected a priori for their role in lipid metabolism/peroxidation. Our findings suggest that SNPs in the APOE/C1 gene region are associated with renal cancer. In two studies, we found that the high risk G-G haplotype, that included SNPs rs8106922 and rs405509, was common among controls in both studies (39%), and conferred a significant 20% increased risk above that observed for the most prevalent low-risk genotype, found in 49% of controls. The APOE protein plays an important role in the cellular uptake of lipoproteins through ligand-receptor interactions with the low-density lipoprotein receptor (LDL) and chylomicron remnant receptors (23). Through this interaction, APOE mediates the uptake and metabolism of lipoproteins and is thought to be a major determinant of blood lipid levels in humans, a precursor to risk factors for renal cancer. In the NOS2A gene, three high-risk haplotypes were identified across introns 5 through 12. Within this region, we selected SNP rs4795067 for replication although it alone was not driving the association observed. Additional genotyping is warranted to identify other causal variants across this region. Moreover, only the combined analysis was sufficiently powered to detect the OR (1.34) observed.
Variation in APOE was previously evaluated for associations with risk factors such as hypertension and body mass index (BMI), and diseases such as biliary tract cancer (24), gall stones (25, 26) and Alzheimer's disease (27,28). In a recent study the APOE variant (IVS1+69C>G; rs440446), the G allele was significantly associated with biliary stones, bile duct cancer, and cancer of the ampulla of Vater (24). The APOE gene has three major isoforms encoded by three alleles (e2, e3, e4) resulting from the presence of two non-synonymous SNPs located within coding regions for amino acids 112 and 158. The allele distribution in most Caucasian populations for the most prevalent e3 allele (Cys112; Arg158) ranges from 70-85%, the e4 allele (Arg112; Arg 158) from 10-20%, and the e2 allele (Cys112; Cys158) from 5-10%. The APOE e4 allele has been associated with increased levels of total and LDL serum cholesterol, whereas the e2 allele was associated with the reverse effect (23). Using these alleles as markers, one study showed an increased risk of gallstones associated with the e4 allele, compared to subjects that had the e3 and e2 alleles (29). In addition to lipid metabolism, this protein also modulates proinflammatory and inflammatory responses (23, 29). It has also been shown to modify risk of other malignancies including breast, colon/rectum, and prostate (30-32). In our study, APOE SNPs were selected to tag regions upstream from and including the SNP rs405509 which is located within the APOE gene promoter. The functional SNP rs405509 that was associated with reduced risk in our studies was recently shown to modify the differential protein binding and transcriptional activity of the APOE gene in HEPG23 liver cells and astrocytoma cells (33,34). This region is in high LD with other SNPs in the APOE gene promoter and is tightly correlated with the location of known codon 112 and 158 variants which define the e2, e3, e4 alleles, that are also thought to be under regulatory control by the regions tagged in this analysis (34).
To the best of our knowledge, this is the first large scale evaluation of key candidate genes involved in lipid metabolism/peroxidation and RCC susceptibility that provide evidence to support that common functional variation at the APOE gene locus may increase susceptibility to RCC. Validation studies in additional independent study populations will be necessary to confirm these findings to exclude the possibility of chance findings.
This work was supported by the Intramural Research Program of the National Institutes of health, National Cancer Institute, division of Cancer Epidemiology and Genetics (Bethesda, MD, USA).