The PON1 gene contains several common single nucleotide polymorphisms, some of which directly affect OP metabolism. In the promoter region, adjacent to a binding site for the transcription factor Sp1 (Deakin et al. 2003), the C-108T polymorphism influences expression of the gene, with the PON1_-108T allele conferring reduced PON1 levels. It is the most influential known polymorphism in the promoter region, contributing 22-25% of variation in PON1 expression in white adults (Brophy et al. 2001; Leviev and James 2000). Relative to PON1_-108CC homozygotes, PON1_-108TT homozygotes have, on average, 33-45% lower enzyme activity as adults (Brophy et al. 2001; Leviev and James 2000) and 63% lower as neonates (Chen et al. 2003). PON1’s OP detoxification activity is also influenced by enzyme structure, and a coding region polymorphism, Q192R, determines whether glutamine (Q) or arginine (R) is present near PON1’s catalytic center (Harel et al. 2004). The two resulting PON1₁₉₂ isoforms hydrolyze some substrates at different rates. In mice, PON1_R192 isoform provides significantly better protection than does PON1_Q192 from chlorpyrifos oxon (Li et al. 2000). The two isoforms provide similar protection with respect to the oxon of diazinon.

Because chlorpyrifos and diazinon are common in children’s environments (Andrew Clayton et al. 2003) and PON1 genotype may influence susceptibility to these OPs (Cole et al. 2003), we examined whether the inefficient PON1 promoter allele (PON1_-108T) or the allele coding for the PON1₁₉₂ isoform that may provide lower chlorpyrifos protection (PON1_192Q) are associated with increased risk of CBT in children.

Subject identification and specimen collection. Institutional review board approvals were received from the Fred Hutchinson Cancer Research Center and Washington State Department of Health (WDOH) before the conduct of this study. Most subjects were drawn from a previous case-control study (Gurney et al. 1996), with additional control subjects randomly selected from similar birth years. Briefly, in the previous study, cases were diagnosed with primary tumors of the brain, cranial nerves, or meninges [International Classification of Diseasesfor Oncology (ICD-O; World Health Organization [WHO] 1976), codes 191.0-192.1] at < 20 years of age in 1984-1991 while residing in the Seattle-Puget Sound region of Washington State covered by a population-based cancer incidence registry affiliated with the National Cancer Institute’s Surveillance, Epidemiology, and End Results Program. Controls from the same counties were identified via random digit dialing (RDD), frequency matched to cases 2:1 by sex and age. Mothers of all participating children [134 (74%) eligible cases and 281 (79%) eligible controls] were interviewed using a structured questionnaire that included questions about home pesticide treatment, defined as chemical treatment of the house for pests such as termites, fleas, ants, cockroaches, or silverfish, during the index pregnancy and/or childhood up to the diagnosis date (cases) or similar reference date (controls).

A subset of these subjects [70 (52%) cases and 160 (57%) controls] were eligible for the present study because the child was born after 1977 and the mother resided in Washington when the child was born, so a dried blood spot (DBS) card potentially remained archived at the WDOH Newborn Screening Program. Blind to case status, staff located cards and clipped a DBS for 66 (94%) eligible cases and 137 (86%) eligible controls. Most (74%) of the unlocated subjects were born either in the earliest, uncatalogued months (14 of 27) or outside a civilian hospital (6 of 27). Case-
control, race/ethnicity, and histologic tumor type proportions from the previous study were preserved in the subset for whom DBS were available. Subjects’ DBS were permanently anonymized (American Society of Human Genetics 1996) before removal from WDOH. DBS from a pilot study that randomly sampled 100 anonymous infants from the same archives and a similar range of birth years (1980-1991) were available to supplement our control group.

DNA extraction and genotyping. DNA extraction and genotyping were conducted at the Center for Ecogenetics and Environmental Health Functional Genomics Laboratory at the University of Washington, blind to case status. Six 3-mm punches were removed from each DBS, with all instrumentation flame sterilized between specimens. DNA was then extracted using the QIAamp DNA Mini Kit (QIAGEN, Valencia, CA) according to the manufacturer’s DBS protocol.

PON1 C-108T and Q192R variants were identified with TaqMan detection system-based assays (respective probes from Integrated DNA Tehcnologies, Coralville, IA, and Applied Biosystems, Foster City, CA). Genotypes were assigned based on relative fluorescence, verified by sequencing as needed. Negative controls and DNA-sequenced positive controls representing each possible genotype were included in each batch of analyses. We included blind duplicate or quadruplicate specimens for 6% of cases and 6% of interviewed controls. All PON1 genotypes were represented, and the results were in complete agreement with the original specimens. In addition, the laboratory repeated the assays for > 10% randomly selected specimens, also in full agreement.

PON1 genotyping was completed for all subjects. One control for whom we collected two equally well-matching DBS was excluded from our CBT-PON1 analyses because PON1 genotype was different for the two possible matches. Thus, 66 cases and 236 controls were available for statistical analysis.

Statistical analysis. Genotype and allele frequencies were tabulated, and chi-square tests were used to check Hardy-Weinberg equilibrium. Using Intercooled Stata (version 8.0; Stata Corp., College Station, TX), we conducted logistic regression to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for CBT in relation to PON1 genotype (Breslow and Day 1980). Because PON1 heterozygotes have an intermediate phenotype (Brophy et al. 2001; Leviev and James 2000), each polymorphism was modeled linearly (0, 1, or 2 PON1_-108T alleles; 0, 1, or 2 PON1_192Q alleles). To allow for possible threshold effects (e.g., any vs. no PON1_R192 isoform), the appropriateness of the linear assumption was verified using the likelihood-ratio test and by comparing modeled risk estimates to those obtained for individual genotypes. Test of trend p-values for CBT and number of PON1_-108T or PON1_192Q alleles were derived from the linear terms in the logistic regression models.

To consider the possible combined effect of the two polymorphisms, we investigated whether risk estimates for one polymorphism depended on the other and calculated CBT risk estimates in relation to PON1_C-108T/Q192R diplotypes and haplotypes. For the latter, we modeled the number of each allele (TQ, TR, CQ, CR) linearly, using all subjects for whom haplotype could be directly inferred or accurately estimated (PHASE, version 2.0.2) (Stephens et al. 2001). We also used PHASE to test whether cases and controls had different haplotype frequencies.

CBT (Gurney et al. 1999) and the PON1_-108T and PON1_192Q alleles (Brophy et al. 2002; Chen et al. 2003) are more common in non-Hispanic whites than in individuals of most other races or ethnicities. To investigate whether population stratification influenced risk estimates, we repeated all analyses restricted to children whose biologic mother and father were both non-Hispanic white. Race/ethnicity was not otherwise included in models because there were few nonwhite or Hispanic subjects, and within this heterogeneous category for which prevalence of PON1 variants varies widely, there were substantial racial/ethnic differences between cases and controls.

Other potential confounders considered were the other PON1 polymorphism (C-108T by Q192R, and vice versa) and the frequency-matching variables (sex and age). These were retained in the models only if ORs or 95% CIs for the PON1 polymorphisms were altered by > 10%, and unless stated, unadjusted risk estimates are presented. We also examined whether CBT-PON1 risk estimates varied by reported home pesticide treatment, farm residence, or parental agricultural occupation, potential indicators of exposure to OPs metabolized by PON1. Statistical interaction on the multiplicative scale was assessed in logistic regression models by the p-value ( = 0.05) of the interaction term, or likelihood-ratio test when a single interaction required multiple terms. To the extent possible, we calculated risk estimates by histologic subtype: astroglial tumors (ICD-O histology codes 9380-9384, 9400-9421, 9424-9442), primitive neuroectodermal tumors (PNETs; 9362, 9470-9473, 9500), and other CBTs, using all controls as the reference group (WHO 1976).

Table 1

Table 2

Subject characteristics. Approximately half of cases had astroglial tumors, with the remainder evenly divided between PNETs and a heterogeneous group of other tumors (Table 1). Most cases (71%) were diagnosed before the age of 5 years. Similar proportions of cases (88%) and controls (91%) were non-Hispanic white; however, among subjects for whom the father’s race/ethnicity was known (63 cases, 135 interviewed controls), proportionally fewer cases (81%) than controls (93%) were born to two non-Hispanic white parents.

Proportionally fewer mothers of cases (31%) than controls (42%) reported that any of the child’s homes had ever been chemically treated for pests during the pregnancy or childhood before the diagnosis/reference date. In this largely urban/suburban region, few subjects had ever lived on a farm (9% cases, 3% controls) or had parents who worked in agriculture (14% cases, 10% controls).

PON1 genotype.PON1 genotype frequencies did not significantly differ from Hardy-Weinberg equilibrium (both p > 0.20). Proportionally more cases (26%) than controls (17%) were homozygous for the inefficient PON1 promoter allele (PON1_-108T), and risk of CBT was nonsignificantly increased with increasing PON1_-108T alleles (for PON1_-108TT, relative to PON1_-108CC: OR = 2.1; 95% CI, 0.9-4.7; for PON1_-108CT, OR = 1.4; 95% CI, 1.0-2.2; p-value for trend = 0.07; Table 2). The association was strongest and statistically significant in relation to the PNET histologic tumor type specifically (for each additional PON1_-108T allele: OR = 2.4; 95% CI, 1.1-5.4; p-value for trend = 0.03, based on 15 PNET cases, including 6 PON1_-108TT homozygotes and 7 heterozygotes; data not shown). According to PON1₁₉₂ genotype, 48% cases and 42% controls had no PON1_R192 isoform (Table 2). Although there was a weak suggestion of increased CBT risk in relation to increasing number of PON1_192Q alleles, CIs were quite wide and the p-value for trend was nonsignificant (for PON1_192QQ, relative to PON1_192RR: OR = 1.5; 95% CI, 0.6-3.4; for PON1_192QR: OR = 1.2; 95% CI, 0.8-1.9; p-value for trend = 0.36). None of the above risk estimates was attenuated when we restricted the analysis to children with two non-Hispanic white parents, nor were they markedly altered when we separately used either the interviewed controls identified through RDD or the anonymous WDOH archive controls as the reference group.

Logistic regression models indicated no interaction (p = 0.75) between these two PON1 polymorphisms. Indeed, we observed a positive association between CBT and the PON1_-108T allele within each PON₁₉₂ genotype, and the estimated risk per PON1_-108T allele was nearly identical in children with (OR = 1.5; 95% CI, 0.9-2.6) or without (OR = 1.4; 95% CI, 0.8-2.5) any PON1_R192 isoform (data not shown). However, with respect to CBT-PON1_Q192R, the possible weak association was absent among PON1_-108TT homozygotes (per PON1_192Q allele relative to PON1_192RR: OR = 1.0; 95% CI, 0.4-2.8.

PON1 diplotype and haplotype. We observed proportionally more cases than controls in each PON1_C-108T/Q192R diplotype with two inefficient PON1 promoters and in the CT/QQ diplotype (no PON1_R192 isoform and only one efficient promoter allele; data not shown). Proportionally fewer cases were represented in each of the other diplotypes, those with two efficient promoters or those with only one efficient promoter but some PON1_R192 isoform.

PON1_C-108T/Q192R haplotype frequencies were not significantly different in cases versus controls (p = 0.09). The haplotype model confirmed the earlier impression that the PON1_192Q allele was not associated with CBT among PON1-_108TT homozygotes: the risk of CBT relative to children with the CR/CR haplotype (homozygous for efficient promotion of PON1_192R isoform) was 1.7 (95% CI, 1.0-3.2) per TQ allele, 1.7 (95% CI, 0.6-4.3) per TR allele, and 1.3 (95% CI, 0.7-2.6) per CQ allele. However, the resulting risk estimates for individual diplotypes were not markedly different from those estimated by simpler models, including one with only a single linear term for each polymorphism.

Table 3

Table 4

PON1 and pesticide exposure indicators. CBT risk was associated with PON1_C-108T only among children whose mothers reported that at least one of the child’s homes had been chemically treated for pests. Relative to PON1_-108CC, the risk per PON1_-108T allele was 2.6 (95% CI, 1.2-5.5), whereas among children whose homes were reportedly not treated, the risk was 0.9 (95% CI, 0.5-1.6; interaction p = 0.03; Table 3). Any suggestion of an interaction between PON1_Q192R and home pesticide treatment was not statistically significant (p = 0.33; Table 4). These observations did not appear to be caused by case-control differences in demographic factors associated with pesticide use/reporting, such as race/ethnicity, maternal education, or smoking. Our ability to examine the CBT-PON1 relation by farm residence or parental agricultural occupation was quite limited, although there was a higher frequency of PON1_-108TT cases among children who had lived on a farm (4 of 6 cases, 0 of 4 controls). Combining all three pesticide exposure indicators, the risk of CBT per PON1_-108T allele was 2.0 (95% CI, 1.03-3.7) among exposed and 1.0 (95% CI, 0.5-1.9) among unexposed (interaction p = 0.15; data not shown.)

This small population-based study suggests that having an inefficient PON1 promoter allele at position -108 is associated with an increased risk of CBT. The observed association was strongest with respect to PNET, the CBT type most consistently associated with farm residence (Bunin et al. 1994; Kristensen et al. 1996). For the most part, CBT was not associated with the PON1_Q192R polymorphism, which determines the enzyme’s structure and thereby detoxification efficiency for some substrates. Our results were similar when we focused on the largest racial/ethnic group in our population, indicating that potential biases related to race/ethnicity were probably not largely responsible for our observations. In addition, risk estimates were fairly resilient to exclusion of either of the two population-based sources of controls. Nevertheless, any association between CBT and PON1_C-108T, and the lack thereof between CBT and PON1_Q192R, must be interpreted with caution.

First, our small numbers of subjects may have hampered our ability to observe a CBT-PON1_Q192R association and simultaneously increased the probability that the apparent association between CBT and PON1_C-108T is a false positive. Furthermore, no prior studies have examined this relationship. Non-Hodgkin lymphoma (Kerridge et al. 2002), multiple myeloma (Lincz et al. 2004), and prostate cancer (Antognelli et al. 2004; Marchesani et al. 2003) have been associated with other PON1 polymorphisms, including Q192R, but to our knowledge this is the first cancer study to consider the C-108T polymorphism of PON1, and the first study to examine the potential role of this enzyme in relation to childhood cancer.

Second, we obtained subjects’ DNA from an indirect source. It is possible that environmental DNA contamination occurred, which would likely bias risk estimates toward null. However, our assays do not simply detect the presence of an allele, but instead rely on relative amounts of each allele compared with sequence-verified laboratory controls to assign one of three genotypes.

Third, although our specimen retrieval rate for cases was very high and did not require cases to have survived, we cannot rule out survival-related selection bias among cases in the study from which they were drawn. However, even if such bias existed and PON1 genotype influenced cancer prognosis, it is unlikely this could account fully for the moderately strong association we observed between CBT and PON1_C-108T.

To the extent that this association is not due to chance or case survival, these results suggest that CBT risk may be inversely related to PON1 enzyme levels. Ideally, fresh blood would have been available and PON1 enzyme levels measured, but the C-108T polymorphism is a significant determinant of PON1 levels in both neonates and pregnant women (Chen et al. 2003). Because PON1 may hydrolyze OPs before they can reach the brain, our results lend support to prior epidemiologic studies that have observed an increased risk of CBT in relation to possible pesticide exposure. OPs target the developing nervous system (Garcia et al. 2002; Johnson et al. 1998), and chlorpyrifos may affect replication and differentiation of glial cells (Garcia et al. 2001).

If OPs do relate to CBT risk, one would expect CBT-PON1 associations to be present mainly among those exposed to chlorpyrifos and/or diazinon. These two OPs were the most common residential insecticides during the study years, and indoor air is an important source of children’s exposure to them (Andrew Clayton et al. 2003). Therefore, it is interesting that we observed an increased risk of CBT in relation to the inefficient PON1_-108T allele only among children whose homes were reportedly treated for insect pests. Still, one cannot discount the possibility that this OP-metabolizing enzyme could protect the brain via its more generic ability to metabolize oxidized lipid molecules. That PON1_Q192R was not associated with CBT seems to underscore this point. However, the relative protection provided by the two resulting PON1₁₉₂ isoforms depends on the OP (Li et al. 2000), and perhaps other factors such as PON1 levels. We had no direct measures of these, nor of the level of exposure to chlorpyrifos and diazinon. Because of our modest number of cases, we also had limited ability to consider other potentially relevant factors, such as diet, age at diagnosis, family history of cancer, and prenatal and childhood exposure to tobacco smoke. For example, there was some indication that the PON1_192Q allele is associated with increased risk of CBT among subjects reportedly not exposed to tobacco smoke, but we could not adequately examine whether this reflects a plausible biologic effect or is simply a spurious observation due to chance or bias related to racial/ethnic differences between cases and controls and smokers and nonsmokers.

The strengths of our study are population-based identification of cases and controls, the use of a DNA source unrelated to case survival, and inclusion of children diagnosed before the residential phase-out of chlorpyrifos and diazinon began. Future studies of CBT and PON1 would benefit from larger sample sizes, more accurate indicators of exposure to chlorpyrifos and diazinon, the addition of other PON polymorphisms, measurement of plasma PON1 activity (e.g., PON1 status) (Costa and Furlong 2002), and genotyping/phenotyping of both children and mothers. It would also be useful to know whether the PON1_C-108T polymorphism is associated with biomarkers of relevance to cancer, such as chromosome aberrations, as has been demonstrated in farmers with respect to PON1_Q192R (Au et al. 1999). Such studies would be worthwhile in light of the possible association we observed between CBT occurrence and the predominant polymorphism in the PON1 promoter region, and because chlorpyrifos remains a leading agricultural insecticide and is detected in or on foods frequently consumed by children (Andrew Clayton et al. 2003; U.S. Department of Agriculture 2004).